1 00:00:03,280 --> 00:00:05,880 Hello and welcome to the Behavioral Design Podcast. 2 00:00:05,960 --> 00:00:08,680 This season we're diving into the intersection of behavioral 3 00:00:08,680 --> 00:00:11,240 science and AI. We want to make the sense of the 4 00:00:11,240 --> 00:00:14,280 state of AI, from understanding how humans interact with 5 00:00:14,280 --> 00:00:17,720 intelligent systems to using AI to do behavioral design itself. 6 00:00:17,920 --> 00:00:21,440 I'm Aline Holsworth, a health tech advisor specializing in AI 7 00:00:21,440 --> 00:00:24,080 and product design. Over the past 15 years, I've 8 00:00:24,080 --> 00:00:26,880 been crafting human centered products with behavioral science 9 00:00:26,920 --> 00:00:29,360 at the core. At Apple, I LED Behavioral 10 00:00:29,360 --> 00:00:32,280 Science for Health AI, designing and launching AI powered 11 00:00:32,280 --> 00:00:34,600 features to help users reach their health goals. 12 00:00:35,320 --> 00:00:37,320 And I'm Samuel Sultzer, your second Co host. 13 00:00:37,560 --> 00:00:40,840 I'm a behavioral strategist specializing in hybrid formation 14 00:00:41,120 --> 00:00:43,360 and designing products that drive long term baby change. 15 00:00:43,840 --> 00:00:47,080 I work with leading tech organizations integrating AI to 16 00:00:47,080 --> 00:00:48,400 scale behavioral design for good. 17 00:00:49,000 --> 00:00:52,360 And I'm also the founder of Baby Bites, a dedicated community on 18 00:00:52,640 --> 00:00:56,440 behavioral science and AI. Quick word on Nuance Behavior 19 00:00:56,440 --> 00:00:59,840 where we help organizations build impactful digital products 20 00:00:59,840 --> 00:01:02,840 using behavioral design. We only take on a few clients at 21 00:01:02,840 --> 00:01:05,400 a time to ensure the highest level of quality for our 22 00:01:05,400 --> 00:01:07,600 tailored evidence based solutions. 23 00:01:08,480 --> 00:01:11,280 If you'd like to become one of our special projects, e-mail us 24 00:01:11,320 --> 00:01:14,920 at hello@nuancebehavior.com or we could call directly on our 25 00:01:14,920 --> 00:01:31,720 website, nuancebehavior.com. It's good to be back, Haley. 26 00:01:32,360 --> 00:01:36,120 Hi, it's very good to be back. Yeah, I can't believe it's been 27 00:01:36,120 --> 00:01:37,760 like 10 months. What happened? 28 00:01:38,080 --> 00:01:42,120 I'm not who's counting What can happen in 10 months? 29 00:01:42,480 --> 00:01:46,880 Yeah, what can happen? What can be born out of that 30 00:01:47,000 --> 00:01:49,960 period? Yes, yes, about 40 weeks or so. 31 00:01:52,640 --> 00:01:55,080 Yeah, it's, it's, it's really good to be back. 32 00:01:55,080 --> 00:01:58,920 I'm honestly super happy to to get back on track with this. 33 00:01:58,920 --> 00:02:03,560 And I'm, yeah, so excited about today as it marks kind of our 34 00:02:03,880 --> 00:02:08,400 return of the podcast, the first episode of Season 4. 35 00:02:08,759 --> 00:02:11,800 And there's so much to get to. We're going to do a little bit 36 00:02:11,800 --> 00:02:13,480 of a season preview. We're going to cover some of the 37 00:02:13,480 --> 00:02:16,320 biggest myths in kind of intersection of AI and 38 00:02:16,320 --> 00:02:19,240 behavioral science. We're going to cover some top AI 39 00:02:19,240 --> 00:02:23,680 terms that any behavioral scientist should know and some 40 00:02:23,680 --> 00:02:27,800 other fun updates and things, but I think we should just jump 41 00:02:27,800 --> 00:02:31,040 straight into it and cover some breaking news. 42 00:02:32,200 --> 00:02:37,080 Yes, it is indeed breaking and and local to you, Sam, Did you 43 00:02:37,080 --> 00:02:41,160 know that very recently, right around the corner from you the 44 00:02:41,640 --> 00:02:44,800 the Nobel Prize in Physics was just awarded? 45 00:02:46,040 --> 00:02:47,920 No, this is actually Full disclosure. 46 00:02:48,280 --> 00:02:51,080 You told me just before the episode that this has happened. 47 00:02:52,000 --> 00:02:52,360 So. So. 48 00:02:52,760 --> 00:02:54,760 Yeah, and we're pretending that I'm telling you now. 49 00:02:55,240 --> 00:02:58,120 I know. And it's funny because yeah, 50 00:02:58,120 --> 00:03:00,800 this is definitely news that maybe I would be expected to 51 00:03:00,800 --> 00:03:02,320 know before because it happened in Stockholm. 52 00:03:02,320 --> 00:03:05,280 But no. Yeah, big news. 53 00:03:05,880 --> 00:03:09,600 Yeah, it's very exciting. And the this year's winners of 54 00:03:09,600 --> 00:03:14,240 the Nobel Prize in Physics are notable for many reasons. 55 00:03:14,240 --> 00:03:17,760 One, because how did they not win the Nobel Prize before now, 56 00:03:17,960 --> 00:03:22,280 but also because it's it's perfectly aligned with our focus 57 00:03:22,320 --> 00:03:25,880 of this season of the podcast. So the winners were John 58 00:03:25,880 --> 00:03:30,920 Hopfield and Jeffrey Hinton, who surely are recognizable names to 59 00:03:30,920 --> 00:03:35,680 all of you, no? Well, maybe, maybe for some. 60 00:03:35,680 --> 00:03:37,240 I think, yeah, for him to, definitely. 61 00:03:37,560 --> 00:03:40,120 Be more. Yeah, he's he's been more on the 62 00:03:40,120 --> 00:03:44,200 pages with many of these things. Is that not always the work that 63 00:03:44,200 --> 00:03:46,760 leads into people being well known, but more of the 64 00:03:46,760 --> 00:03:49,400 controversy. So the fact that he recently 65 00:03:49,400 --> 00:03:53,040 stepped down from his work at Google and talked about stepping 66 00:03:53,040 --> 00:03:57,440 down in order to be more outspoken about the the risks 67 00:03:57,440 --> 00:04:02,520 and ideas of of AII think maybe made him more well known 68 00:04:02,600 --> 00:04:04,400 recently. But he's been known, yeah. 69 00:04:04,960 --> 00:04:08,440 I feel like the the media machine jumped on that, I think. 70 00:04:08,520 --> 00:04:12,480 I think the reality is probably a lot less exciting than the 71 00:04:12,480 --> 00:04:13,960 headlines would lead you to believe. 72 00:04:14,840 --> 00:04:16,480 Yeah. But I mean, just to hear from 73 00:04:16,480 --> 00:04:19,120 you like, why do you think like this is exciting or why do you 74 00:04:19,120 --> 00:04:22,280 think this is kind of important to highlight their work and why 75 00:04:22,280 --> 00:04:25,160 are they deserving of being recognized here in Stockholm? 76 00:04:25,840 --> 00:04:29,520 Yeah, Stockholm specifically, yeah. 77 00:04:30,480 --> 00:04:34,880 So I think that the award for sure was all about their work in 78 00:04:34,880 --> 00:04:38,640 machine learning and, but specifically in creating or and 79 00:04:38,640 --> 00:04:42,480 developing and making progress in artificial neural networks 80 00:04:42,880 --> 00:04:45,240 and. And this is important because 81 00:04:45,440 --> 00:04:49,200 most of the, the machine learning and artificial 82 00:04:49,200 --> 00:04:53,200 intelligence is really based on neural networks and, and this 83 00:04:53,200 --> 00:04:56,640 structure for processing information and deep learning 84 00:04:56,640 --> 00:04:58,720 and so on. Yeah, you could almost call it 85 00:04:58,720 --> 00:05:01,480 the foundation of machine learning the the work that 86 00:05:01,480 --> 00:05:03,640 they've done over the years. Yeah. 87 00:05:03,640 --> 00:05:07,640 And I think it is very fascinating where we're so quick 88 00:05:07,640 --> 00:05:12,040 to talk about the things that AI can do and and advancement and 89 00:05:12,040 --> 00:05:14,400 so on, but it's easier to forget kind of what it builds upon. 90 00:05:14,480 --> 00:05:18,920 We'll get into talking more maybe about LLMS and and the the 91 00:05:18,920 --> 00:05:23,440 recent hyper on some of the innovative AI where it feels 92 00:05:23,440 --> 00:05:26,240 like they are weird humans because they're basically been 93 00:05:26,240 --> 00:05:30,400 trained on human knowledge and human human things, but they're 94 00:05:30,400 --> 00:05:32,160 not humans. They also they're like weird 95 00:05:32,160 --> 00:05:35,360 humans in a way. The the work from John and and 96 00:05:35,400 --> 00:05:38,920 Joffrey, it goes even deeper in terms of looking at the brain 97 00:05:38,920 --> 00:05:43,200 stuff mimicking the human brain. Yeah, yeah. 98 00:05:43,200 --> 00:05:48,680 And, and it's so interesting, I sometimes I talk to computer 99 00:05:48,680 --> 00:05:54,440 scientists who who are, are very familiar with artificial neural 100 00:05:54,440 --> 00:05:58,200 networks and not at all familiar with the biological neural 101 00:05:58,200 --> 00:06:00,640 networks. So how our brains are structured 102 00:06:00,640 --> 00:06:03,600 and how our neurons talk to each other and how they fire and send 103 00:06:03,840 --> 00:06:07,640 information with their chemical and electrical signals. 104 00:06:08,440 --> 00:06:12,560 And so I, I've often found myself more often than you might 105 00:06:12,560 --> 00:06:15,240 think, in a position where I have to explain that no, 106 00:06:15,240 --> 00:06:21,000 artificial neural networks are while while they mimic the brain 107 00:06:21,000 --> 00:06:24,920 and have many similarities with the structure of the brain, they 108 00:06:24,920 --> 00:06:27,800 are not the same as biological neural networks. 109 00:06:27,800 --> 00:06:30,560 In fact, there are some really, really important differences 110 00:06:30,560 --> 00:06:33,720 between the two. So I thought maybe it could be 111 00:06:33,720 --> 00:06:38,400 interesting to talk a little bit about those differences and what 112 00:06:38,400 --> 00:06:43,040 actually does an artificial neural network look like versus 113 00:06:43,040 --> 00:06:48,040 a biological neural network. Yeah, I I would love for you to 114 00:06:48,040 --> 00:06:50,320 explain that in some ways I would trust you more than me. 115 00:06:50,320 --> 00:06:54,000 How would you? In this hypothetical scenario? 116 00:06:54,000 --> 00:06:56,480 They had to explain it to someone sitting across here. 117 00:06:56,800 --> 00:06:59,840 How would you distinguish between the artificial and the 118 00:06:59,840 --> 00:07:02,320 biological? Yeah, well, for the sake of our 119 00:07:02,400 --> 00:07:05,520 behavioral scientist listeners, I'll do a little bit of the 120 00:07:05,520 --> 00:07:09,040 explainer on the the machine learning part as well. 121 00:07:09,040 --> 00:07:13,200 So it, I think it, it sort of helps to to visualize what this 122 00:07:13,200 --> 00:07:15,520 looks like. And of course, it's much, much, 123 00:07:15,520 --> 00:07:19,360 much more complicated than you could even begin to describe. 124 00:07:19,400 --> 00:07:23,440 But if if you think about layers and layers and layers, think of 125 00:07:23,440 --> 00:07:26,320 it in three sections, OK, you have your inputs, you have your 126 00:07:26,320 --> 00:07:28,520 hidden layers, and then you have your output layers. 127 00:07:28,720 --> 00:07:31,800 So if you're like reading this as you would on a screen from 128 00:07:31,800 --> 00:07:34,040 left to right, you have your inputs on the left. 129 00:07:34,040 --> 00:07:37,800 And then you have many, many, many layers and connections in, 130 00:07:38,240 --> 00:07:42,240 in the hidden portion, which is often many more layers than you 131 00:07:42,240 --> 00:07:44,240 could even imagine. And so that's where the 132 00:07:44,240 --> 00:07:47,280 complexity of artificial neural networks comes from. 133 00:07:47,480 --> 00:07:51,200 And then you have your output layers all the way to the right, 134 00:07:51,360 --> 00:07:56,560 and then data flows from left to right as you would be reading. 135 00:07:58,040 --> 00:08:01,840 Now, learning, especially in the case of of deep learning, 136 00:08:01,920 --> 00:08:04,920 happens through what machine learning scientists call back 137 00:08:04,920 --> 00:08:08,880 propagation. So as data are flowing in from 138 00:08:08,880 --> 00:08:13,880 left to right, there's each of these nodes in the layers has a 139 00:08:13,880 --> 00:08:17,240 little weight where they're they're making calculations and 140 00:08:17,280 --> 00:08:20,680 math is happening. I think like the the most high 141 00:08:20,680 --> 00:08:24,520 level way to think about it is like the data flows through, but 142 00:08:24,520 --> 00:08:28,680 in order to learn we have to send information back through. 143 00:08:28,680 --> 00:08:32,320 So going back through from the right to the left and this is 144 00:08:32,400 --> 00:08:35,440 back propagation. This is the really like what the 145 00:08:35,559 --> 00:08:37,720 the learning part of machine learning is. 146 00:08:37,720 --> 00:08:41,440 And when that's happening, the weights in the middle are being 147 00:08:41,440 --> 00:08:44,280 adjusted. So if something say say there 148 00:08:44,280 --> 00:08:48,560 was a calculation at some point where it was said multiply by .3 149 00:08:48,840 --> 00:08:52,360 and then you get to the end in the in the output and you 150 00:08:52,360 --> 00:08:54,760 learned, Oh well there was some error there. 151 00:08:54,760 --> 00:09:00,880 Our predicted value of 1 * .3 instead of being the prediction 152 00:09:00,880 --> 00:09:04,760 of .3 it was actually .5 OK, well now you have a .2 error and 153 00:09:04,760 --> 00:09:08,280 you send that back through the system and then in the future 154 00:09:08,280 --> 00:09:10,200 iteration you have a corrected weight. 155 00:09:12,920 --> 00:09:15,560 Is that too complicated? No, I think that's good. 156 00:09:15,720 --> 00:09:20,480 And like this work on artificial neural networks is what they've 157 00:09:20,480 --> 00:09:26,560 been kind of recognize for. And I guess again, it will be 158 00:09:26,560 --> 00:09:30,080 interesting then to contrast it briefly with the biological 159 00:09:30,440 --> 00:09:32,800 neural network. How are they kind of similar, 160 00:09:32,800 --> 00:09:36,440 different, Yeah. So if all of that sounded very 161 00:09:36,800 --> 00:09:40,760 complex, the first thing to know about human brains is human 162 00:09:40,760 --> 00:09:44,920 brains are way, way, way, way more complex. 163 00:09:46,360 --> 00:09:49,640 Oh my gosh. So in our brains, we have 164 00:09:49,800 --> 00:09:52,520 billions of neurons. They're connected through these 165 00:09:52,960 --> 00:09:57,400 through our synapses, and our synapses make trillions of 166 00:09:57,400 --> 00:09:59,440 connections number. Big. 167 00:09:59,560 --> 00:10:03,240 Big, big numbers more, more, more than a, a machine learning 168 00:10:03,240 --> 00:10:04,960 model. And I'm actually not gonna go 169 00:10:04,960 --> 00:10:08,800 into the complexity of neurons because that there is, there's 170 00:10:08,800 --> 00:10:12,680 so much to unpack there, right? But I think that an important 171 00:10:12,960 --> 00:10:18,080 take away is that whereas artificial neural networks can, 172 00:10:18,080 --> 00:10:22,520 as a result of their structure, be very good, often better than 173 00:10:22,520 --> 00:10:27,920 biological neural networks at at very specific tasks with lots 174 00:10:27,920 --> 00:10:31,360 and lots and lots of data. The strength of our biological 175 00:10:31,360 --> 00:10:35,160 neural networks is that they're highly dynamic. 176 00:10:35,760 --> 00:10:39,800 They we, we have synaptic plasticity where we can change 177 00:10:39,880 --> 00:10:44,200 the strength of our connections changes very quickly and often 178 00:10:44,200 --> 00:10:49,200 with very little data. So in, in the case of AI, you 179 00:10:49,200 --> 00:10:51,360 need lots and lots and lots of data and you have very little 180 00:10:51,360 --> 00:10:55,000 flexibility. And with brains, you, you might 181 00:10:55,000 --> 00:10:58,920 have one or two observations and you learn very, very quickly 182 00:10:58,920 --> 00:11:03,760 with with very little data. And this, I'm biased as a 183 00:11:03,760 --> 00:11:08,440 behavioral scientist, but makes me believe more and more just 184 00:11:08,440 --> 00:11:12,200 how unbelievably incredible our brains are. 185 00:11:12,200 --> 00:11:16,160 You cannot replicate a brain with math, not yet. 186 00:11:16,640 --> 00:11:21,360 And we're not close at all. The fundamental structure of 187 00:11:21,360 --> 00:11:26,200 artificial neural networks is such that they do not have this 188 00:11:26,240 --> 00:11:30,680 ability to to be flexible. Yeah, I think very fun thing 189 00:11:30,840 --> 00:11:34,320 recently has been seeing and experiencing kind of that 190 00:11:34,880 --> 00:11:39,680 there's been actually a lot of people have noticed that have 191 00:11:40,000 --> 00:11:43,080 increased appreciation for the complexity and the beauty of the 192 00:11:43,080 --> 00:11:46,480 brain and how complexity is. I've actually had several people 193 00:11:47,040 --> 00:11:51,760 that are coming more from some interest in wanting to better 194 00:11:51,760 --> 00:11:55,800 understand machine learning or AI that I have started to read 195 00:11:55,800 --> 00:11:58,120 more about cognitive processes and the brain. 196 00:11:58,120 --> 00:12:01,440 There's so much work that just looks at like just in all of the 197 00:12:01,640 --> 00:12:05,040 things that the brain can do, tries to mimic it tries to in 198 00:12:05,040 --> 00:12:07,080 some ways learn from how we work. 199 00:12:07,200 --> 00:12:10,400 And, and I think that's quite, Yeah, beautiful. 200 00:12:11,680 --> 00:12:15,360 The brain is beautiful and really, if you think machine 201 00:12:15,360 --> 00:12:19,520 learning models can be a black box, tell me how consciousness 202 00:12:19,520 --> 00:12:26,320 arises from our meat machines. It's just magic. 203 00:12:27,000 --> 00:12:30,800 No, but I do love that quote from from Marvin Minsky. 204 00:12:30,800 --> 00:12:33,400 He's one of the original AI thinkers. 205 00:12:33,400 --> 00:12:36,400 You might call him that. He says the brain is merely a 206 00:12:36,400 --> 00:12:41,120 meat machine. Yeah, that is an amazing quote. 207 00:12:41,200 --> 00:12:46,840 I I want to put that on a yeah AT shirt or some form of a 208 00:12:46,840 --> 00:12:50,880 hoodie or a cap or something. But talk more about that quote. 209 00:12:51,400 --> 00:12:52,840 Yeah. I love it. 210 00:12:52,880 --> 00:12:57,360 I, I think, well, I think what's so interesting about it is to 211 00:12:57,360 --> 00:12:59,720 think about our brains in terms of machines, right? 212 00:12:59,720 --> 00:13:03,440 And, and not only how we use the brain to model artificial 213 00:13:03,440 --> 00:13:08,000 intelligence systems, right, but we then use AI sort of the, the 214 00:13:08,000 --> 00:13:11,320 converse and, and technology more generally to then think 215 00:13:11,320 --> 00:13:15,000 about the mind. So going both ways, we both mold 216 00:13:15,000 --> 00:13:18,480 our computers after ourselves like like our neural networks, 217 00:13:18,760 --> 00:13:21,680 but then we then we can't help being influenced by our 218 00:13:21,680 --> 00:13:28,360 technology in return. That actually that's something 219 00:13:28,360 --> 00:13:32,920 that Sherry Turkle has been studying for something like 40 220 00:13:32,920 --> 00:13:36,880 years, right? So basically the the advent of 221 00:13:36,880 --> 00:13:39,400 the computer age. Are you familiar with Sherry 222 00:13:39,400 --> 00:13:41,360 Turkle? No, no, tell me where. 223 00:13:41,640 --> 00:13:44,080 Yeah, let me, let me tell you just very briefly. 224 00:13:44,320 --> 00:13:50,400 So Sherry is a sociologist at MIT and she is an incredible 225 00:13:50,400 --> 00:13:52,840 person. I, I actually just Full 226 00:13:52,840 --> 00:13:55,800 disclosure, I just finished her memoir, The Empathy Diaries. 227 00:13:55,800 --> 00:14:00,400 So I'm now an expert on Sherry. So I, I won't shame you for not 228 00:14:00,400 --> 00:14:05,360 knowing, but what was really cool about Sherry and her work 229 00:14:05,360 --> 00:14:09,080 was that she captured this shift really well. 230 00:14:09,080 --> 00:14:12,320 Computers were just coming about really like the personal 231 00:14:12,320 --> 00:14:14,800 computer in, in around the 1970s. 232 00:14:14,800 --> 00:14:18,600 Like this is like it was starting large scale, small 233 00:14:18,600 --> 00:14:22,960 scale, then large scale adoption where people were moving from 234 00:14:22,960 --> 00:14:25,880 thinking about humans as rational animals. 235 00:14:25,880 --> 00:14:28,800 That was sort of what defined the human was you're an animal, 236 00:14:28,800 --> 00:14:32,600 but with rationality to to thinking of them as emotional 237 00:14:32,600 --> 00:14:34,600 machines. So in reference to the machine 238 00:14:34,600 --> 00:14:37,280 now, not in reference to animals. 239 00:14:37,960 --> 00:14:42,560 And really since then, I mean, my entire life for sure, we've 240 00:14:42,560 --> 00:14:47,040 had this mantra that humans are unique, where these special, 241 00:14:47,040 --> 00:14:51,920 unique creatures because we can express emotions and empathy. 242 00:14:51,920 --> 00:14:55,240 That's what's like really special about humans. 243 00:14:55,920 --> 00:15:00,280 Well, consciousness for sure. But, but, but especially when we 244 00:15:00,280 --> 00:15:05,080 think about emotion and now we see our AI systems maybe not 245 00:15:05,080 --> 00:15:09,600 feeling emotion, but certainly expressing emotion, expressing 246 00:15:09,600 --> 00:15:14,240 empathy, making, tricking, you could say people into feeling 247 00:15:14,240 --> 00:15:19,520 like they can rely on this thing or, or whatever you want to call 248 00:15:19,520 --> 00:15:22,160 it. What does that mean? 249 00:15:22,640 --> 00:15:26,920 But that's a big question. But I, I know we have to have to 250 00:15:26,920 --> 00:15:30,480 get to our actual episode so we can save that big, big topic for 251 00:15:30,760 --> 00:15:34,600 another day. But I do think that Sherry's 252 00:15:34,600 --> 00:15:38,760 work exploring this interplay between humans and technology 253 00:15:39,120 --> 00:15:42,800 really sort of captures the essence of what we're trying to 254 00:15:42,800 --> 00:15:45,080 do with this season of the podcast. 255 00:15:45,640 --> 00:15:48,040 Yeah, so you're predicting that she's going to be the next one 256 00:15:48,080 --> 00:15:50,880 for a Nobel Prize? I hope so. 257 00:15:51,240 --> 00:15:53,800 I hope so that would that would be due. 258 00:15:54,400 --> 00:15:57,560 Yeah, no, but it's so true. And I I like that kind of frame 259 00:15:57,560 --> 00:16:03,520 of of it and AI just briefly on what you're saying, I think it 260 00:16:03,520 --> 00:16:07,600 is interesting that pair the original definition of Turing 261 00:16:07,600 --> 00:16:13,200 test in some ways like the current AI models, they are 262 00:16:13,400 --> 00:16:16,960 already succeeding. They are able to trick anyone to 263 00:16:16,960 --> 00:16:19,520 believe that they're real human beings if you're chatting with 264 00:16:19,520 --> 00:16:22,000 them. Like there's no way to to really 265 00:16:22,000 --> 00:16:24,560 know. But at the same time, as you 266 00:16:24,560 --> 00:16:27,920 say, it's also two things can be true. 267 00:16:27,920 --> 00:16:29,680 That can be true in terms of Turing tests. 268 00:16:29,680 --> 00:16:33,400 The traditional one at least has been kind of accomplished, but 269 00:16:34,000 --> 00:16:41,240 we still probably separate a pretty big part of that to when 270 00:16:41,320 --> 00:16:43,840 a large langu of mole saying, if you ask them, is that OK? 271 00:16:44,040 --> 00:16:46,440 Can you feel feelings? And do you feel sad? 272 00:16:46,440 --> 00:16:50,200 And like, if it would say like, yes, I feel sad and feel lonely 273 00:16:50,480 --> 00:16:53,760 and so on, we would picked out with a bit of a pinch of salt in 274 00:16:53,760 --> 00:16:55,600 terms of what what actually that means. 275 00:16:55,600 --> 00:16:59,920 And, and we won't kind of I I guess there was some of these 276 00:16:59,920 --> 00:17:03,160 news a couple years ago with some people being like, Oh my 277 00:17:03,160 --> 00:17:07,240 God, I asked to shut sorry, I asked AI this and they said yes. 278 00:17:07,240 --> 00:17:10,000 And I'm, I'm afraid. Well, I'm pretty I I think we 279 00:17:10,000 --> 00:17:11,800 have a little more nuanced approach these days. 280 00:17:12,160 --> 00:17:14,079 Yeah. And I think you and I know quite 281 00:17:14,079 --> 00:17:18,079 a bit about this intersection of behavioral science and AI, but 282 00:17:18,079 --> 00:17:21,839 we're learning too. And, and honestly, so much is 283 00:17:21,839 --> 00:17:27,040 changing every every week, every day, that just keeping up on 284 00:17:27,200 --> 00:17:29,760 what the literature and what's published at this intersection 285 00:17:30,080 --> 00:17:32,640 is could be a full time job for us. 286 00:17:33,280 --> 00:17:35,920 Maybe someday it will be. But for now, for now, we'll 287 00:17:35,920 --> 00:17:39,520 we'll keep our side jobs. And I, I guess that brings us to 288 00:17:39,520 --> 00:17:40,920 the season. Do you want to tell the audience 289 00:17:40,920 --> 00:17:43,520 a little bit of what the plan is for for this season? 290 00:17:44,200 --> 00:17:47,840 Yeah, we're, we're really trying to cover everything from 291 00:17:48,280 --> 00:17:51,760 understanding how humans interact with intelligent 292 00:17:51,760 --> 00:17:56,880 systems to using AI to to do behavioral design itself. 293 00:17:56,880 --> 00:17:59,400 So how do we use it in our work? Yeah. 294 00:17:59,400 --> 00:18:03,320 And I today I feel very related to I felt about 6:00 or so years 295 00:18:03,320 --> 00:18:07,040 ago now when I initially started have a weekly where I felt with 296 00:18:07,040 --> 00:18:10,120 behavioral design, which is really, really hard to stay on 297 00:18:10,120 --> 00:18:13,400 top of what's going on. And as a petitioner to kind of, 298 00:18:14,560 --> 00:18:19,600 yeah, just feel like you are in the loop of like, what are other 299 00:18:19,600 --> 00:18:21,280 people doing? How are other people thinking? 300 00:18:21,560 --> 00:18:24,320 I think right now it's a really triggered time for anyone as a 301 00:18:24,320 --> 00:18:27,600 be able scientist to really make sense of maybe both from 302 00:18:28,000 --> 00:18:30,040 identity point of view, like who am I today? 303 00:18:30,680 --> 00:18:33,960 But also just feeling kind of more on solid ground when it 304 00:18:33,960 --> 00:18:37,040 comes to like AI, Like what are useful things for us as 305 00:18:37,040 --> 00:18:40,800 petitioners to know? What are things that are already 306 00:18:40,800 --> 00:18:44,480 people are doing and what can we learn from each other and, and 307 00:18:44,480 --> 00:18:48,520 really kind of do that openly and and collectively? 308 00:18:48,520 --> 00:18:50,280 I think that's kind of where we're coming from, I guess. 309 00:18:50,880 --> 00:18:54,200 Yeah, I think, I think we can do the work of sifting through 310 00:18:54,400 --> 00:18:58,920 everything that's put out there, not everything, but many things 311 00:18:59,720 --> 00:19:01,640 and share what's relevant and interesting. 312 00:19:02,160 --> 00:19:05,200 That's that's really the goal. So we can be as useful as 313 00:19:05,200 --> 00:19:06,720 possible. Yeah. 314 00:19:06,720 --> 00:19:10,800 And I think from my end, if I'm completely honest, putting all 315 00:19:10,800 --> 00:19:14,960 the cards on the table, I personally would love through 316 00:19:14,960 --> 00:19:21,120 decision to have a better personal argument for used the 317 00:19:21,320 --> 00:19:24,160 business case as well around payable sense and AI, 'cause I 318 00:19:24,160 --> 00:19:28,120 have a lot of opinions around the importance of understanding 319 00:19:28,120 --> 00:19:33,200 that AI is not a form of panacea or silver bullet or magic wand 320 00:19:33,200 --> 00:19:35,920 to solve a lot of things that it's being used for. 321 00:19:36,280 --> 00:19:40,200 And we've already had some, we should teach this, but like some 322 00:19:40,200 --> 00:19:43,320 really good recordings of, of guests episodes where we've 323 00:19:43,320 --> 00:19:47,720 already explored the importance of thoughtfully kind of marrying 324 00:19:47,720 --> 00:19:50,440 behavioral science and AI at the right places and, and the really 325 00:19:50,440 --> 00:19:54,600 big benefits that gives. But I'm really excited to to 326 00:19:54,600 --> 00:20:00,720 better formulate that argument and that thesis of sort so that 327 00:20:00,880 --> 00:20:01,880 that doesn't get lost. Yeah. 328 00:20:02,760 --> 00:20:05,840 I think it's easy to find yourself on either of those 329 00:20:05,840 --> 00:20:09,720 extremes where you're jumping on the bandwagon with everyone, 330 00:20:09,720 --> 00:20:13,200 everyone and you're all AI everywhere. 331 00:20:14,120 --> 00:20:16,920 Or you maybe you find yourself on the other end of the spectrum 332 00:20:16,920 --> 00:20:19,760 where you say everyone is doing AI. 333 00:20:20,080 --> 00:20:23,640 Like I just being totally dismissive and I will be honest, 334 00:20:23,640 --> 00:20:28,640 I, I have been that person on that end of the spectrum at at 335 00:20:28,640 --> 00:20:30,880 points in my life. And I don't think that's the 336 00:20:30,880 --> 00:20:33,760 right place to be either. I think that finding the right 337 00:20:33,760 --> 00:20:37,840 balance and understanding where is AI useful? 338 00:20:38,240 --> 00:20:40,480 Where is it actually the best tool? 339 00:20:41,280 --> 00:20:46,680 But yeah, how do we actually use this most effectively and just 340 00:20:46,680 --> 00:20:50,760 understand its impact on our science and on society? 341 00:20:50,760 --> 00:20:53,000 I think that is just absolutely essential. 342 00:20:56,600 --> 00:20:59,720 Hey everyone, this is Maddie, your AI producer of this 343 00:20:59,720 --> 00:21:02,320 episode. I just thought this was a good 344 00:21:02,320 --> 00:21:04,000 opportunity to jump in real quick. 345 00:21:04,520 --> 00:21:07,160 We really value listener involvement this season and we 346 00:21:07,160 --> 00:21:09,680 would love to hear your thoughts, ideas, and reflections 347 00:21:09,680 --> 00:21:12,920 as you listen along. So if you've got anything on 348 00:21:12,920 --> 00:21:16,200 your mind, topics you want us to cover, guests you'd love to hear 349 00:21:16,200 --> 00:21:18,680 from, or anything about the intersection of behavioral 350 00:21:18,680 --> 00:21:21,000 science and AI, feel free to reach out. 351 00:21:21,320 --> 00:21:24,240 You can e-mail us at podcast@habitweekly.com. 352 00:21:24,640 --> 00:21:26,760 We read all your messages and and really want to make this 353 00:21:26,760 --> 00:21:29,240 season as relevant and valuable as possible for you. 354 00:21:30,320 --> 00:21:33,680 Again, e-mail us at podcast@habitweekly.com. 355 00:21:34,080 --> 00:21:35,920 Now let's get back to Sam and Elaine. 356 00:21:39,680 --> 00:21:43,480 Wow. So yeah, I'm really excited for 357 00:21:43,480 --> 00:21:49,080 this season and maybe also to share, is it useful to maybe 358 00:21:49,080 --> 00:21:51,680 talk about where we're coming from, like both you and I and 359 00:21:52,080 --> 00:21:53,320 all of this? Yeah, yeah, yeah. 360 00:21:53,560 --> 00:21:57,720 So we did mention that we have some expertise in AI or this 361 00:21:57,720 --> 00:22:00,960 intersection between behavioral science and AI. 362 00:22:00,960 --> 00:22:05,200 So I think it's useful to sort of lay down our cards and say 363 00:22:05,320 --> 00:22:10,440 here's why you should trust us. So I, my life's work has really 364 00:22:10,440 --> 00:22:14,440 been in helping people reach the goals that they have often their 365 00:22:14,440 --> 00:22:17,200 health goals, not always their health goals, but you know, how 366 00:22:17,200 --> 00:22:19,960 do we, how do we help people exercise when they want to 367 00:22:19,960 --> 00:22:23,080 exercise and so on. And so I've been doing that, the 368 00:22:23,280 --> 00:22:25,760 digital health space for a long, long time. 369 00:22:26,680 --> 00:22:30,680 And then it may be in the past five years or so, really got 370 00:22:30,680 --> 00:22:35,320 interested in AI, specifically in terms of reaching these same 371 00:22:35,320 --> 00:22:37,280 goals. How do we, how do, how do we 372 00:22:37,280 --> 00:22:39,360 leverage new technologies to do so? 373 00:22:39,720 --> 00:22:42,160 So then I joined Apple. I thought this would be a, a 374 00:22:42,280 --> 00:22:46,920 good way of achieving this goal. I LED behavioral science for its 375 00:22:46,920 --> 00:22:50,000 health AI team. So that was a, a really fun 376 00:22:50,000 --> 00:22:54,480 experience working on high impact features, working with 377 00:22:54,520 --> 00:22:58,200 fitness, plus really, really working at this intersection of 378 00:22:58,320 --> 00:23:02,040 AI and behavioral science. And one thing that I should 379 00:23:02,040 --> 00:23:06,920 mention is that before I entered Apple, I really had very, very 380 00:23:06,920 --> 00:23:10,240 little trading and machine learning and, and the artificial 381 00:23:10,240 --> 00:23:13,080 intelligence. So I, I sort of took it upon 382 00:23:13,080 --> 00:23:18,080 myself to dive head first into this whole world and Oh my gosh, 383 00:23:18,120 --> 00:23:24,000 like what a world it is. It's so, so wide and deep. 384 00:23:24,400 --> 00:23:29,680 So I, I read books, went to conferences, took courses, like, 385 00:23:29,920 --> 00:23:32,400 I even learned how to code a bit myself. 386 00:23:32,400 --> 00:23:35,560 I mean, don't ask me to actually produce something, but let me 387 00:23:35,560 --> 00:23:40,120 just say that I, I feel that now that I've gone through that 388 00:23:40,120 --> 00:23:44,280 experience, I am, I am a much more humble person when it comes 389 00:23:44,280 --> 00:23:46,840 to realizing how much I don't know. 390 00:23:47,360 --> 00:23:54,040 And, and I also, I feel the a little smirk every time that I 391 00:23:54,040 --> 00:23:58,560 see people on the Internet talking about AI without really 392 00:23:58,560 --> 00:24:00,320 knowing the first thing about it. 393 00:24:00,960 --> 00:24:04,400 And so I feel like I was that person and I can see what 394 00:24:04,400 --> 00:24:06,760 they're doing, but I can also see right through what they're 395 00:24:06,760 --> 00:24:11,440 doing. So I think that, yeah, our our 396 00:24:11,440 --> 00:24:15,360 goal of this season is definitely going to be to expose 397 00:24:15,360 --> 00:24:19,080 you and ourselves to the people who know quite a lot about 398 00:24:19,080 --> 00:24:21,080 artificial intelligence and behavioral science. 399 00:24:21,800 --> 00:24:25,000 So. But with that, Sam, tell me 400 00:24:25,000 --> 00:24:30,040 about yourself and what's what's your cred, your street cred in 401 00:24:30,040 --> 00:24:33,320 this world of? Artificial intelligence, it is 402 00:24:33,320 --> 00:24:35,720 hard to compete with being the behavioral science lead at 403 00:24:35,840 --> 00:24:38,120 health AI at Apple. I think it's hard to. 404 00:24:39,520 --> 00:24:43,120 Yeah, it's not a competition. No, but I think it's the same, 405 00:24:43,120 --> 00:24:44,720 but different in some ways. And I think that's why it's 406 00:24:44,720 --> 00:24:49,120 always a little bit interesting because I had quite of a 407 00:24:49,120 --> 00:24:56,280 different entry when it came to engaging with AI, where this was 408 00:24:56,920 --> 00:25:01,240 when the GPT 3A model from Open AI came out. 409 00:25:01,240 --> 00:25:03,560 So that was I think a year or so before. 410 00:25:03,600 --> 00:25:07,160 Yeah, somewhat old school, but it was really the model that 411 00:25:07,360 --> 00:25:09,720 where you start to engage with it and you're like, wow, this is 412 00:25:10,440 --> 00:25:12,160 pretty cool. Like you can actually do quite a 413 00:25:12,160 --> 00:25:16,120 lot of things with it. And, and so it's I think a year 414 00:25:16,120 --> 00:25:20,280 or so maybe a little bit more before chat TPT came out and 415 00:25:20,400 --> 00:25:23,920 very randomly I was working with some client projects where that 416 00:25:23,920 --> 00:25:27,040 started to be used and and, and started to be considered and, 417 00:25:27,400 --> 00:25:29,920 and I was part of really leading some of that and, and working 418 00:25:29,920 --> 00:25:34,440 with that. And so I, I was tempted to 419 00:25:34,440 --> 00:25:37,640 become this kind of people that do the red teaming and and so 420 00:25:37,640 --> 00:25:40,080 forth, testing these models. I didn't end up going that far 421 00:25:40,120 --> 00:25:43,400 in, in kind of working directly with developing the models, but 422 00:25:43,400 --> 00:25:47,000 I, I feel like I quite quickly got a very intimate experience 423 00:25:47,000 --> 00:25:51,240 with, with working with them. And so that's a lot of the work 424 00:25:51,240 --> 00:25:53,800 that I've been doing. So maybe less on, I would say 425 00:25:54,200 --> 00:26:00,280 you definitely come from having much more of AI would say rich 426 00:26:00,600 --> 00:26:03,360 machine learning, understanding of, of AI. 427 00:26:03,800 --> 00:26:08,120 Where a lot of my work has come in on more of the generative AI 428 00:26:08,120 --> 00:26:11,760 side where it's been trying to use these models to, to make 429 00:26:11,760 --> 00:26:16,080 sense in some form of the kind of multimodal aspect of either 430 00:26:16,440 --> 00:26:20,320 kind of making sense of kind of images through using AI or 431 00:26:20,320 --> 00:26:24,480 creating text or understanding text or creating personalized 432 00:26:24,480 --> 00:26:28,680 interventions and and so on with these models and using prompt 433 00:26:28,680 --> 00:26:32,240 engineering and fine tuning and dragging and some of these 434 00:26:32,240 --> 00:26:34,600 things that is coming out to kind of improve the consistency 435 00:26:34,600 --> 00:26:37,520 and reliability of them. So. 436 00:26:37,680 --> 00:26:42,320 Well, and just to give you a little compliment here, I would 437 00:26:42,320 --> 00:26:47,280 say you have mastered the art of prompt engineering, even in just 438 00:26:47,280 --> 00:26:50,640 our our work doing like background research and that 439 00:26:50,640 --> 00:26:52,880 sort of thing. I will throw something into 440 00:26:52,920 --> 00:26:56,000 ChatGPT and and share my results with you. 441 00:26:56,000 --> 00:26:59,760 And then you go and hack away at it for a couple minutes and you 442 00:26:59,760 --> 00:27:03,440 come back with the like shiny emerald version of what I had. 443 00:27:03,440 --> 00:27:07,800 It's just I've been awestruck by what you can do just just by 444 00:27:07,800 --> 00:27:13,040 your experience tinkering around and really exploring these these 445 00:27:13,400 --> 00:27:16,360 getting your hands dirty with with these technologies. 446 00:27:17,000 --> 00:27:19,600 Yeah, yeah. And then I, I will definitely be 447 00:27:19,600 --> 00:27:22,760 honest with that. I have done both to that degree, 448 00:27:22,760 --> 00:27:25,800 but also I think recently the last year or so, I've been 449 00:27:25,800 --> 00:27:30,920 really excited about the agentic side of what is happening in 450 00:27:30,920 --> 00:27:33,560 terms of what we're, we're going to be exploring that a bit. 451 00:27:33,560 --> 00:27:36,280 And when we mean when we say agents, what we mean is 452 00:27:36,280 --> 00:27:43,320 basically that you have some form of AI set up to be a a 453 00:27:43,920 --> 00:27:47,160 multi purpose tool of sort where it can do multiple things for 454 00:27:47,160 --> 00:27:49,600 you depending on the task you're kind of asking it for you. 455 00:27:50,240 --> 00:27:52,280 Essentially automating your workflow. 456 00:27:52,280 --> 00:27:55,000 Is that what you're talking? About yeah, for example, yeah, 457 00:27:55,000 --> 00:27:57,360 it's a lot of automation comes into kind of like using some 458 00:27:57,360 --> 00:28:00,520 form of agentic things. And, and so yeah, I've really 459 00:28:00,520 --> 00:28:07,040 become yeah, very deep into that space of like using AI for 460 00:28:07,040 --> 00:28:09,680 automation purposes. And, and really, I think I find 461 00:28:09,680 --> 00:28:12,040 it so interesting to understand just when it comes to 462 00:28:12,040 --> 00:28:16,280 automation, how again, the human aspect were it's easier 463 00:28:16,280 --> 00:28:18,080 something done to automate things because even though you 464 00:28:18,080 --> 00:28:20,680 kindly could automate it, there's always this explicit 465 00:28:20,680 --> 00:28:23,200 things that needs to be done, but also implicitly things we 466 00:28:23,200 --> 00:28:25,520 don't think about what we're doing and, and how how that 467 00:28:25,520 --> 00:28:28,000 plays into things. So, so yeah, that's a long, long 468 00:28:28,000 --> 00:28:32,880 with a way of saying that I, yeah, I, I've had my fair share 469 00:28:32,880 --> 00:28:36,680 of exposure to, to AI, but as you, I feel also very humble. 470 00:28:37,120 --> 00:28:40,440 I feel so much so much that I feel still that I, I want to 471 00:28:40,440 --> 00:28:41,920 know more about and understand better. 472 00:28:41,920 --> 00:28:45,520 And, and I think that's, that's also where I've really enjoyed 473 00:28:45,600 --> 00:28:47,320 our private discussions. And I feel like I've learned so 474 00:28:47,320 --> 00:28:49,880 much from from you. We're kind of privately talking 475 00:28:49,880 --> 00:28:51,880 about some of these things. So, so yeah, I'm really excited 476 00:28:51,880 --> 00:28:55,080 to have this season up and running again and having this 477 00:28:55,080 --> 00:29:01,440 recorded conversations too, and. Yeah, I, I feel like I have to 478 00:29:01,440 --> 00:29:05,800 mention how our, our current professional lives intersect 479 00:29:05,800 --> 00:29:09,200 with this as well. Because in, in the in the 480 00:29:09,200 --> 00:29:13,160 meantime, since I guess since we've since we've last released 481 00:29:13,160 --> 00:29:16,800 an episode, we have created a company together with some 482 00:29:16,800 --> 00:29:22,000 friends nuanced behavior And at that company we are working on 483 00:29:22,000 --> 00:29:26,400 advising on AI driven tech. So this is exactly what we do, 484 00:29:26,400 --> 00:29:30,720 sort of injecting behavioral design and behavioral into 485 00:29:31,040 --> 00:29:36,120 products that use AI. They don't have to use AI, but 486 00:29:36,120 --> 00:29:38,240 many of them do. Yeah. 487 00:29:38,240 --> 00:29:39,320 And I think that's what's interesting. 488 00:29:39,440 --> 00:29:43,160 We, we ended up calling this nuanced behavior and it's easy 489 00:29:43,160 --> 00:29:47,000 to kind of find yourself using the word nuanced quite often. 490 00:29:47,000 --> 00:29:51,800 But it is where we're coming from where we have a lot of 491 00:29:51,800 --> 00:29:54,040 experience when it comes to applied rural science. 492 00:29:54,520 --> 00:29:58,080 And it's been really interesting to to meet a lot of clients that 493 00:29:59,040 --> 00:30:03,720 maybe have a really strong capability when it comes to some 494 00:30:03,720 --> 00:30:07,280 of these aspects, potentially even in AI in some cases, but 495 00:30:07,280 --> 00:30:10,640 also helping them get the most value from the behavioral 496 00:30:10,640 --> 00:30:13,880 science side. Firstly, when understanding the 497 00:30:13,880 --> 00:30:18,240 human, like how do we really support humans in leading better 498 00:30:18,240 --> 00:30:20,800 lives? And whether it's sustainability 499 00:30:20,800 --> 00:30:25,240 or in health or in in more financial related context, how 500 00:30:25,240 --> 00:30:29,680 can we can we both with the technological understanding and 501 00:30:29,680 --> 00:30:32,800 capability, but still not forgetting that in the end, we 502 00:30:32,800 --> 00:30:35,240 have to understand the human side of it and we have to make 503 00:30:35,240 --> 00:30:36,920 the most of that. I think that's been really, 504 00:30:36,920 --> 00:30:40,680 really fun to to complement with an income really from that side 505 00:30:40,680 --> 00:30:43,080 as well. Yeah, absolutely. 506 00:30:43,560 --> 00:30:49,800 So given that this is our intro episode to the season, we 507 00:30:49,800 --> 00:30:55,240 thought that it might be useful to share some overarching themes 508 00:30:55,240 --> 00:30:59,400 about AI and behavioral science, share some terminology that's 509 00:30:59,400 --> 00:31:03,400 going to come up throughout the season and really kind of get 510 00:31:03,400 --> 00:31:05,640 into what are the myths out there? 511 00:31:05,640 --> 00:31:10,280 We can, we can sort of debunk those those myths and set the 512 00:31:10,280 --> 00:31:12,760 record straight. And I thought that the, the 513 00:31:12,760 --> 00:31:16,640 version of this that we could do is AI1O1 for behavioral 514 00:31:16,640 --> 00:31:19,160 scientists. And, and I want to do this in 515 00:31:19,160 --> 00:31:23,600 the spirit of a, a course that that existed back when I was an 516 00:31:23,600 --> 00:31:26,640 undergrad at Reed College. It was called physics for 517 00:31:26,640 --> 00:31:31,320 philosophers. And you can just tell by the by 518 00:31:31,320 --> 00:31:34,400 the name of the course, this is not your hardcore physics. 519 00:31:34,720 --> 00:31:38,560 So this is our version of AI for behavioral scientists, just to 520 00:31:38,560 --> 00:31:41,320 give kind of a general what's what. 521 00:31:42,440 --> 00:31:45,960 And I think there are some really basic basics to know 522 00:31:45,960 --> 00:31:51,200 about AI for, for the very beginner and, and in terms of 523 00:31:51,200 --> 00:31:55,200 just how we think about AI, of course, we've already spoken a 524 00:31:55,280 --> 00:31:57,560 little bit already about about neural networks. 525 00:31:57,560 --> 00:32:04,440 But if you think of I, I think at their very core AI systems 526 00:32:04,480 --> 00:32:07,880 are and in and in particular machine learning systems. 527 00:32:07,880 --> 00:32:12,440 I'm not even going to get into teasing those two apart, but if 528 00:32:12,440 --> 00:32:15,920 you if you think of them as super powered prediction 529 00:32:15,920 --> 00:32:20,760 machines, right, you are taking data and from that data finding 530 00:32:20,760 --> 00:32:25,440 patterns in the data, often massive amounts of data, and 531 00:32:25,440 --> 00:32:29,200 then categorizing that data, classifying it and making a 532 00:32:29,200 --> 00:32:30,360 prediction. Yeah. 533 00:32:30,360 --> 00:32:33,920 I think that's a really helpful way to simplify it and really 534 00:32:33,920 --> 00:32:38,960 think in my asset super powered prediction machine also reminds 535 00:32:38,960 --> 00:32:41,680 me of this conversation we had at some point with Lisa Feldman 536 00:32:41,680 --> 00:32:48,040 Barrett about emotion and so on where she talked about us and 537 00:32:48,040 --> 00:32:52,040 how we engage and understand our emotion, this kind of prediction 538 00:32:52,840 --> 00:32:55,120 machines. And then it ties into this whole 539 00:32:55,360 --> 00:33:00,240 thing that that episode started with and and that of course we 540 00:33:00,240 --> 00:33:02,160 are also prediction machine. But I when we're talking about 541 00:33:02,200 --> 00:33:05,960 artificial system, that's, yeah, narrow, narrow pitch machines. 542 00:33:05,960 --> 00:33:08,920 Narrow pitch machines? Not so. 543 00:33:08,920 --> 00:33:14,840 Good at common sense or information transfer or or 544 00:33:14,840 --> 00:33:18,160 really generalizing at a conceptual level that that's a 545 00:33:18,160 --> 00:33:23,800 human thing. For now, SIM, what do you think 546 00:33:23,840 --> 00:33:27,480 are the term the absolutely essential terms that our 547 00:33:27,480 --> 00:33:33,200 listeners have to know in order to understand any conversation 548 00:33:33,200 --> 00:33:36,080 about AI? Where should we start? 549 00:33:36,160 --> 00:33:38,720 Yeah. So let's start sharing maybe 550 00:33:38,720 --> 00:33:42,880 three key terms that I think is helpful to have heard of and 551 00:33:42,880 --> 00:33:48,640 have some understanding around. So I think let's start with 552 00:33:48,760 --> 00:33:54,400 large language models or LLMS. So when people talk about 553 00:33:54,400 --> 00:33:59,120 degenerative AI or Gen. AI, in some ways, they're often 554 00:33:59,120 --> 00:34:03,880 referring to large language models, since these models are 555 00:34:03,880 --> 00:34:09,480 what typically drive any type of generative AI content 556 00:34:09,480 --> 00:34:14,840 generation. And so when it comes to LLMS, 557 00:34:15,000 --> 00:34:20,120 the more popular ones or the, I would say the ones that maybe 558 00:34:20,120 --> 00:34:24,719 even some of our listeners have been using or testing, obviously 559 00:34:24,719 --> 00:34:29,960 you have open AIS GPT 4. And important to understand is 560 00:34:29,960 --> 00:34:32,880 that these models, you should have two components. 561 00:34:32,880 --> 00:34:38,000 They have one maybe shot version, which is like shot GPT 562 00:34:38,639 --> 00:34:43,040 and so on. But they also have AAPI 563 00:34:43,040 --> 00:34:46,320 component, meaning that they can be used directly within others 564 00:34:46,840 --> 00:34:50,719 softwares or in other products. And so when it comes to opening 565 00:34:50,719 --> 00:34:53,920 the eye, a lot of people think about shot GPT, but they 566 00:34:53,920 --> 00:34:58,640 underestimate the fact that their underlying GPT 4 O, for 567 00:34:58,640 --> 00:35:02,680 example, API is really what they're often times interacting 568 00:35:02,680 --> 00:35:06,880 with in many, many places outside of directly shut GPT. 569 00:35:07,520 --> 00:35:11,560 And so then we also have Meta who have their Llama model and 570 00:35:11,560 --> 00:35:15,760 that's quite popular for more kind of open source use cases 571 00:35:16,120 --> 00:35:20,040 and Entropics Cloud or their various cloud models. 572 00:35:20,560 --> 00:35:24,000 And they are probably like the biggest direct competitor right 573 00:35:24,000 --> 00:35:27,560 now to open AI in terms of exactly how they're trying to 574 00:35:27,560 --> 00:35:31,520 structure their their models. And so they're all similar in 575 00:35:31,520 --> 00:35:36,240 that they all generate content and they've been trained on 576 00:35:36,760 --> 00:35:39,200 insane amount of data. Like they've been trained on so 577 00:35:39,200 --> 00:35:44,360 much data from everywhere from Wikipedia to Reddit. 578 00:35:44,680 --> 00:35:53,160 And it's, yeah, it's basically a large degree of whatever we have 579 00:35:53,520 --> 00:35:57,320 available to us from whatever has been recorded within human 580 00:35:57,320 --> 00:36:01,880 knowledge online is being used and has been used to train many 581 00:36:01,880 --> 00:36:05,720 of these models. And so when we talk about them 582 00:36:05,720 --> 00:36:09,360 being kind of weird humans in some ways, that's why. 583 00:36:09,360 --> 00:36:12,920 It's because they are trained on things that we have written on 584 00:36:13,080 --> 00:36:16,920 Wikipedia or Reddit or in books and and so on. 585 00:36:17,560 --> 00:36:26,240 And so overall they are designed to again create text, but 586 00:36:26,240 --> 00:36:30,880 they've also become multi modal in recent years, meaning that 587 00:36:31,240 --> 00:36:35,240 they can often times handle images, audio and even video in 588 00:36:35,240 --> 00:36:38,320 some cases. And so that's why they're also 589 00:36:38,320 --> 00:36:41,200 becoming kind of increasingly powerful because, you know, you 590 00:36:41,200 --> 00:36:46,880 can use the same model to basically look at what you see 591 00:36:46,880 --> 00:36:48,320 in front of you. Like you can take a photo of 592 00:36:48,320 --> 00:36:51,760 what you see and you can ask a model what it, what do I see? 593 00:36:52,040 --> 00:36:55,520 And then it can give you a response in text or an audio 594 00:36:55,520 --> 00:36:59,040 form. And so that can be extremely 595 00:36:59,040 --> 00:37:02,720 powerful in many contexts. And we see now with the 596 00:37:02,720 --> 00:37:07,600 versatility that these models are actually able to perform 597 00:37:07,600 --> 00:37:11,760 about 50% of tasks done by knowledge workers. 598 00:37:12,360 --> 00:37:15,880 It's kind of estimated. Obviously, this is when the LLMS 599 00:37:15,880 --> 00:37:20,320 are set up correctly and the tasks are like explicitly 600 00:37:20,320 --> 00:37:22,440 defined in a very, very clear and simple way. 601 00:37:23,840 --> 00:37:27,200 So, so yeah, they're capable of everything from summarizing to 602 00:37:27,200 --> 00:37:32,240 generating code to even creating personalized interventions and 603 00:37:32,280 --> 00:37:36,840 and so on. And what kind of limits their 604 00:37:36,840 --> 00:37:40,480 power is often times based on the input they're given. 605 00:37:41,200 --> 00:37:44,720 And that brings us to prompt engineering, which is the 606 00:37:44,720 --> 00:37:48,120 process of guiding these models to generate useful and 607 00:37:48,120 --> 00:37:51,560 consistent results. So for example, if you simply 608 00:37:51,560 --> 00:37:55,560 ask the model, like, hey, write me an e-mail, you will probably 609 00:37:55,560 --> 00:37:58,000 get something very generic in return. 610 00:37:58,440 --> 00:38:06,680 But if you provide the in the prompt some clear context, like 611 00:38:07,000 --> 00:38:12,000 write a formal e-mail to a new client introducing our services 612 00:38:12,000 --> 00:38:14,720 and so on. And then when it comes to prompt 613 00:38:14,720 --> 00:38:17,560 engineering, often times it goes much, much, much deeper. 614 00:38:17,800 --> 00:38:22,520 It's often times looks at the idea of providing multiple 615 00:38:22,520 --> 00:38:25,440 examples of the output that you want. 616 00:38:25,920 --> 00:38:32,520 So multi kind of shot prompt or non non zero shot prompts. 617 00:38:32,880 --> 00:38:36,440 So it can also include some form of chain of thought prompting 618 00:38:36,440 --> 00:38:40,480 basically where you're kind of trying to get the model to a 619 00:38:41,160 --> 00:38:46,800 review and reason and and think about what is trying to provide 620 00:38:46,800 --> 00:38:49,240 before actually gives it as an output. 621 00:38:49,840 --> 00:38:54,320 And so hopefully when done properly, you'll get something 622 00:38:54,320 --> 00:38:57,000 much more relevant this way. And I feel like a lot of people, 623 00:38:57,000 --> 00:39:01,480 I guess, get disappointed because they don't realize that 624 00:39:01,480 --> 00:39:05,440 even though these models are very powerful, they still need 625 00:39:05,440 --> 00:39:08,520 clear instructions. And we often talk about the idea 626 00:39:08,520 --> 00:39:12,240 of kind of having 1000 AI interns because that's what 627 00:39:12,240 --> 00:39:15,120 often feels like is that they can do a lot of things. 628 00:39:15,120 --> 00:39:19,080 And they are like, you know, with scale, you can do so much 629 00:39:19,240 --> 00:39:21,280 at same time. You have to keep things simple 630 00:39:21,280 --> 00:39:25,000 and be very clear about the context, and vague prompts 631 00:39:25,040 --> 00:39:28,520 honestly lead to vague results or generic results and so on. 632 00:39:29,320 --> 00:39:32,680 And that's oftentimes the first step to work consistency. 633 00:39:33,080 --> 00:39:36,280 And there's always more advanced kind of steps to take. 634 00:39:36,720 --> 00:39:39,480 We won't get into this too much today, but you can use things 635 00:39:39,480 --> 00:39:46,080 like RAG retrieval, augmented generation, or RLAHF, meaning 636 00:39:46,200 --> 00:39:48,880 reinforcement learning with human feedbacks and so on. 637 00:39:48,880 --> 00:39:51,320 There's a lot of things that could come into play to fine 638 00:39:51,320 --> 00:39:56,200 tune or to improve or to get better kind of content that it 639 00:39:56,200 --> 00:40:00,600 retrieves for its output. But yeah, we'll, we'll, we'll 640 00:40:01,480 --> 00:40:04,040 want to get into those. For now, we'll probably go cover 641 00:40:04,040 --> 00:40:08,920 them more in this season, but yeah, finally the last term so I 642 00:40:08,920 --> 00:40:12,040 covered. Large language models the role 643 00:40:12,040 --> 00:40:15,040 of prompt engineering and then AI agents. 644 00:40:15,080 --> 00:40:18,240 So AI agents are autonomous systems assigned to perform 645 00:40:18,240 --> 00:40:21,040 tasks without needing constant supervision. 646 00:40:21,360 --> 00:40:25,720 So one example is having some form of customer service chat 647 00:40:25,720 --> 00:40:28,240 bot that handles some form of routine queries. 648 00:40:29,440 --> 00:40:32,480 Another one is setting up an AI agent to automatically draft 649 00:40:32,480 --> 00:40:38,000 emails for you. And so the difference here with 650 00:40:38,000 --> 00:40:41,080 an agent is that you can set it up so that it knows how to kind 651 00:40:41,080 --> 00:40:45,160 of perform certain types of writing styles. 652 00:40:45,160 --> 00:40:50,080 So it knows when to adjust the tone and style based on the 653 00:40:50,080 --> 00:40:53,520 recipient. So that when you draft an e-mail 654 00:40:53,520 --> 00:40:56,800 to a colleague, that is different to the e-mail that's 655 00:40:56,800 --> 00:40:59,000 drafted for a client or your partner and so on. 656 00:40:59,280 --> 00:41:02,280 So that's increasingly what my work has been around, you know, 657 00:41:02,280 --> 00:41:04,920 working with these kind of AI agents combined with, you know, 658 00:41:04,920 --> 00:41:09,360 prompt in earring and so on. And they can really create some 659 00:41:09,360 --> 00:41:14,840 powerful, often times automated systems or when when you talk 660 00:41:14,840 --> 00:41:18,240 about automating workflows, they become extremely powerful. 661 00:41:19,560 --> 00:41:25,520 And yeah, the these are the three key concepts that I guess 662 00:41:25,520 --> 00:41:28,640 I wanted to then start with. So we've covered large language 663 00:41:28,640 --> 00:41:31,760 models, a little bit of intro of what they are and some of the 664 00:41:31,760 --> 00:41:35,400 different models out there, the importance of prompt engineering 665 00:41:35,400 --> 00:41:38,600 and how it's just not the same as just writing something into 666 00:41:38,600 --> 00:41:40,800 the model. It's about kind of ensuring 667 00:41:40,800 --> 00:41:46,920 consistency and then how you can scale some of this work when 668 00:41:46,920 --> 00:41:49,800 you're bringing AI agents into to play. 669 00:41:50,080 --> 00:41:54,760 So together they enable us to kind of optimize and scale and, 670 00:41:54,760 --> 00:41:58,560 and automate the tasks in a way that I think, you know, just a 671 00:41:58,560 --> 00:42:03,080 few years ago was completely unthinkable and, and something 672 00:42:03,080 --> 00:42:05,520 we never imagined would be possible today. 673 00:42:07,480 --> 00:42:10,400 So yeah, but there's obviously much more to talk about here. 674 00:42:10,440 --> 00:42:15,920 And so maybe Eileen, you can share something more maybe on 675 00:42:15,920 --> 00:42:19,360 recommender systems or I briefly mention reinforcement learning. 676 00:42:19,360 --> 00:42:22,640 And yeah, what would you want to share with with people? 677 00:42:24,000 --> 00:42:26,680 Yeah, I would love to. I think there are really three 678 00:42:26,680 --> 00:42:31,720 major categories. Recommender systems is 1, 679 00:42:32,000 --> 00:42:34,200 Reinforcement learning is another. 680 00:42:34,480 --> 00:42:37,960 And then I feel like it's easy to forget because we we don't 681 00:42:37,960 --> 00:42:42,560 see as much of this in the at the intersection of AI and 682 00:42:42,560 --> 00:42:44,440 behavioral science. But there's also just our sort 683 00:42:44,440 --> 00:42:50,360 of classical machine learning where you have a bunch of data 684 00:42:50,680 --> 00:42:56,520 and here you say you train your model on a million cat pictures 685 00:42:56,520 --> 00:43:00,240 and then you have some new data and you want to learn is, is 686 00:43:00,240 --> 00:43:02,560 this photo of a cat or a not cat? 687 00:43:03,160 --> 00:43:08,960 You can with fairly fairly good accuracy, understand with many 688 00:43:08,960 --> 00:43:12,280 of these classic machine learning models, whether a photo 689 00:43:12,280 --> 00:43:16,520 is a cat or not cat. And I'm, I'm borrowing from 690 00:43:16,560 --> 00:43:21,920 Cassie Kozikov's famous cat, not cat example because I really 691 00:43:21,920 --> 00:43:24,840 admire her work. But but that's, that's sort of 692 00:43:24,840 --> 00:43:27,960 like the most basic version of a of a classic ML model. 693 00:43:28,760 --> 00:43:31,480 The the other categories that you mentioned, recommender 694 00:43:31,480 --> 00:43:35,760 systems, we talk a lot about recommender systems in our 695 00:43:35,760 --> 00:43:39,080 episode with Carrie Morwich. And so, so you'll have a, a 696 00:43:39,080 --> 00:43:41,920 recommender systems one O 1 in there. 697 00:43:42,520 --> 00:43:45,920 So you'll wait for that launch. And then of course, we have 698 00:43:45,920 --> 00:43:50,120 reinforcement learning, which we really dive into with both Amy 699 00:43:50,120 --> 00:43:55,360 Bucher and Susan Murphy, and we really look at the potential for 700 00:43:55,360 --> 00:44:00,160 using reinforcement learning within behavioral science as a 701 00:44:00,160 --> 00:44:02,280 as a method for doing behavioral science. 702 00:44:03,080 --> 00:44:07,720 So yeah, that covers some useful terminology and we'll attach 703 00:44:07,720 --> 00:44:11,920 this in the show notes as well. We can refer to if, if in doubt, 704 00:44:12,320 --> 00:44:13,800 hopefully that's going to be a little helpful. 705 00:44:14,640 --> 00:44:17,680 But then we also wanted to cover a bit of myths because I think 706 00:44:18,320 --> 00:44:21,080 we both feel like there's quite a lot of those going around when 707 00:44:21,080 --> 00:44:23,760 it comes to talking about AI and especially like this 708 00:44:23,760 --> 00:44:25,280 intersection of behavioral science and AI. 709 00:44:25,760 --> 00:44:31,400 And I think 1 myths or slash like oversimplification that I 710 00:44:31,400 --> 00:44:39,560 feel like I hear a lot is just really equating intelligence in 711 00:44:39,560 --> 00:44:41,680 the same way. Like you have a large language 712 00:44:41,680 --> 00:44:45,640 model that is taking an IQ test and then it's very easy to say, 713 00:44:45,640 --> 00:44:48,600 oh, it's as intelligent as a human being. 714 00:44:48,600 --> 00:44:52,280 So that means that it's as human level intelligence. 715 00:44:53,040 --> 00:44:55,720 And that is true. But I think a lot of people also 716 00:44:55,720 --> 00:44:57,800 don't extend that to say that it's human like. 717 00:44:58,520 --> 00:45:02,520 And I think again, it's really, really, really, really important 718 00:45:02,760 --> 00:45:08,680 to not let us may be seduced or confused or made us to think 719 00:45:08,680 --> 00:45:12,720 that they are one of the same because they are again, really, 720 00:45:12,920 --> 00:45:17,840 really different. I think as I'm seeing the data 721 00:45:17,840 --> 00:45:21,520 is coming out, how impressive some of the models are now at 722 00:45:21,520 --> 00:45:25,320 performing certain tasks, at providing certain types of 723 00:45:25,320 --> 00:45:29,560 expertise. It certainly is leveling or 724 00:45:29,560 --> 00:45:32,440 exceeding human level intelligence in some areas in. 725 00:45:33,280 --> 00:45:35,520 Some. Areas in some areas, in some 726 00:45:35,600 --> 00:45:40,400 areas and it's still a very, very, very big difference to 727 00:45:40,400 --> 00:45:45,400 what is AGI like what, what does it look like, but it we're still 728 00:45:45,400 --> 00:45:47,640 so. Far say what AGI means for our. 729 00:45:48,400 --> 00:45:50,200 Yeah. So artificial general 730 00:45:50,200 --> 00:45:52,560 intelligence, you actually referenced the study of narrow 731 00:45:52,560 --> 00:45:54,800 intelligence because you got things, important things to 732 00:45:54,800 --> 00:45:58,360 really hit home that we're still in a very narrow set of 733 00:45:58,360 --> 00:46:01,680 intelligence where, OK, these large language models can 734 00:46:01,680 --> 00:46:03,560 operate and do a lot of different tasks and a lot of 735 00:46:03,560 --> 00:46:06,520 different things, but they're still very narrow. 736 00:46:06,520 --> 00:46:10,840 That's why we often talk about having 1000 AI interns and this 737 00:46:10,840 --> 00:46:15,520 idea of generalized intelligence where they can highly and 738 00:46:16,800 --> 00:46:20,960 reliably perform in a lot of general contexts and really 739 00:46:20,960 --> 00:46:24,920 understand what we know as human beings have like human like 740 00:46:24,920 --> 00:46:27,400 intelligence. It's still far away. 741 00:46:27,760 --> 00:46:32,360 And you're right, it is extremely seductive when you see 742 00:46:32,360 --> 00:46:38,800 that oh ChatGPT passed the bar exam or ChatGPT outperforms 743 00:46:38,840 --> 00:46:44,640 physicians in XYZ. But you know, then the next 744 00:46:44,640 --> 00:46:47,760 second fails. A very basic cognitive 745 00:46:47,760 --> 00:46:51,240 reflection task. Or are you familiar with the 746 00:46:51,360 --> 00:46:55,280 Nick Bostrom's paper clip? Example, yeah, coming from 747 00:46:55,280 --> 00:46:57,320 effective ultrasound, that's like one of the classic. 748 00:46:57,720 --> 00:47:00,560 Yeah. So this thought experiment is, 749 00:47:01,000 --> 00:47:05,320 say you give an AIA task and it is maximize the production of 750 00:47:05,320 --> 00:47:10,400 paper clips, then maybe the AI with access to all the resources 751 00:47:10,400 --> 00:47:13,040 that it needs can maximize paper clips. 752 00:47:14,440 --> 00:47:16,520 And if that's your goal, maybe that's a good thing. 753 00:47:16,520 --> 00:47:20,400 But of course, there are always unintended consequences. 754 00:47:20,400 --> 00:47:25,480 And when, while maximizing paper clips, you, you might find that 755 00:47:25,760 --> 00:47:28,440 there that there are actually other goals that you want to 756 00:47:28,440 --> 00:47:33,360 have in mind as well, such as not killing humans while you're 757 00:47:33,360 --> 00:47:36,560 at it. And and so then you might tell 758 00:47:36,560 --> 00:47:41,760 the AI, OK, but in, in order to maximize paper clips, please 759 00:47:41,760 --> 00:47:43,760 also don't kill humans. OK? 760 00:47:43,760 --> 00:47:46,520 So like it's, it's given that direct order, it'll do that. 761 00:47:47,160 --> 00:47:50,520 But then you think about all of the other indirect ways that you 762 00:47:50,520 --> 00:47:57,040 might kill humans by destroying the environment by by by 763 00:47:57,960 --> 00:48:00,960 excavating all of the resources that you need in order to 764 00:48:01,040 --> 00:48:03,280 maximize paper clips. And you can sort of go on and on 765 00:48:03,280 --> 00:48:04,800 and on with all of these edge cases. 766 00:48:05,160 --> 00:48:12,120 And you can tell an AI system to directly do all of these things. 767 00:48:12,320 --> 00:48:18,160 But the the contrast is for the human, you don't you, you it's, 768 00:48:18,360 --> 00:48:20,320 it's so, it's common sense, right? 769 00:48:20,320 --> 00:48:26,160 It's so obvious that you don't want to deforest the entire 770 00:48:26,160 --> 00:48:29,560 planet in order to maximize paper clips that that that's not 771 00:48:29,560 --> 00:48:31,480 even something you would have to consider saying. 772 00:48:31,720 --> 00:48:35,960 However, if your only goal as an AI system is to maximize paper 773 00:48:35,960 --> 00:48:38,720 clips, then you might you know, you don't have that 774 00:48:38,720 --> 00:48:40,920 consideration. You don't have that human 775 00:48:40,920 --> 00:48:45,480 element of of common sense to guide your decision making. 776 00:48:45,480 --> 00:48:48,600 You only have the inputs that you're given. 777 00:48:49,800 --> 00:48:52,720 Yeah, and I I mentioned this and I think that's speaks to a lot 778 00:48:52,720 --> 00:48:56,040 of people's experience where I think it's a little bit sad in 779 00:48:56,040 --> 00:48:58,160 some ways as well. But like when they for the first 780 00:48:58,160 --> 00:49:01,240 time they open shat GBT and ask it something and they just get 781 00:49:01,240 --> 00:49:04,200 like a really shitty response because these assets something 782 00:49:04,200 --> 00:49:07,080 very generic and simple and we get some kind of generic and 783 00:49:07,080 --> 00:49:10,480 simple back. That's me if you're describing 784 00:49:10,480 --> 00:49:12,160 me. OK, OK. 785 00:49:13,440 --> 00:49:17,600 But in that case it is easy to feel like, OK, this is shit, but 786 00:49:17,600 --> 00:49:20,720 it again is because these models and and so they're currently 787 00:49:20,720 --> 00:49:22,800 quite far away from real understanding the context that 788 00:49:22,800 --> 00:49:25,880 we're in. And we we're doing our best to 789 00:49:25,880 --> 00:49:29,840 kind of kind of communicate and, and make the context come across 790 00:49:29,840 --> 00:49:33,440 in an efficient way, but it's still still very narrow in in 791 00:49:33,440 --> 00:49:37,360 nature. So yeah. 792 00:49:37,440 --> 00:49:39,200 OK. So what, on your end, do you 793 00:49:39,200 --> 00:49:42,400 feel similarly about as a myth? All right, so we have human 794 00:49:42,400 --> 00:49:47,840 level is does not equal human. Like that's our myth #1 and you 795 00:49:47,840 --> 00:49:51,640 just mentioned Gen. AI, so I wanna throw in a myth 796 00:49:51,640 --> 00:49:56,680 that I think exists about Gen. AI, which is that it boosts 797 00:49:56,680 --> 00:50:00,120 creativity. And this is this is a really fun 798 00:50:00,120 --> 00:50:04,280 one for me because some research recently came out that explored 799 00:50:04,280 --> 00:50:12,720 exactly this looking at OK, it's have have some users ask them to 800 00:50:13,000 --> 00:50:17,640 create a story and either give them access to ChatGPT or not. 801 00:50:18,120 --> 00:50:23,560 And then just objectively assess the, the, the products that they 802 00:50:23,560 --> 00:50:26,440 come up with, how creative is the story that they write? 803 00:50:26,760 --> 00:50:33,480 And it's really interesting that giving access to ChatGPT does 804 00:50:33,480 --> 00:50:39,000 increase creativity on average. So across the board, if you, if 805 00:50:39,000 --> 00:50:42,840 you rate the creativity of these stories, it seems like, yeah, 806 00:50:42,840 --> 00:50:46,880 OK, that that's a that's a plus. But then if you actually drill 807 00:50:46,880 --> 00:50:52,880 down into the the stories, if you look at people who were more 808 00:50:52,880 --> 00:50:58,760 creative at at the start, the sort of your high skill workers, 809 00:50:58,760 --> 00:51:01,560 they really didn't benefit from ChatGPT at all. 810 00:51:01,800 --> 00:51:07,320 This lift was really can be contributed to helping the those 811 00:51:07,320 --> 00:51:12,720 who were not very creative or or or not very skilled in writing 812 00:51:12,920 --> 00:51:17,280 in the 1st place. So you actually find that that 813 00:51:17,320 --> 00:51:21,000 the ChatGPT is much more helpful for for less skilled workers. 814 00:51:21,000 --> 00:51:24,760 This is this is kind of a another myth in itself. 815 00:51:25,960 --> 00:51:30,640 But another side effect of this actually is that even though 816 00:51:30,960 --> 00:51:36,240 the, the, the stories were more creative overall for the, for 817 00:51:36,240 --> 00:51:39,840 the individual users, if you look at the content in a 818 00:51:39,880 --> 00:51:44,320 collective sense, the diversity of stories is actually reduced. 819 00:51:44,320 --> 00:51:49,480 So everyone is kind of getting the same ideas from Chachi BT 820 00:51:49,480 --> 00:51:52,040 and they're all converging on similar stories. 821 00:51:52,720 --> 00:51:57,280 And that to me is not a good thing that that that is really 822 00:51:57,280 --> 00:52:02,080 just making the the content in our world much more homogeneous 823 00:52:02,080 --> 00:52:08,560 and much less interesting. Yeah, and I can speak a little 824 00:52:08,560 --> 00:52:12,720 bit just based on just having a partner who's a motion designer 825 00:52:12,720 --> 00:52:16,320 and and artist coming from an art background. 826 00:52:17,120 --> 00:52:21,200 It's really like I, I see the frustration when she's trying to 827 00:52:21,760 --> 00:52:25,280 use some of these tools because like she either has the option 828 00:52:25,280 --> 00:52:28,400 to do it herself in a full control and rely on her own 829 00:52:28,720 --> 00:52:32,240 unique creativity and so on. Or it's just kind of like almost 830 00:52:32,960 --> 00:52:37,760 tie an hour behind her back and be kind of reliant on a shitty 831 00:52:37,760 --> 00:52:41,000 version of that through some of the quote UN quote creative 832 00:52:41,000 --> 00:52:43,880 tools that can't really do things as well as she can. 833 00:52:44,320 --> 00:52:46,000 And where she doesn't have full control. 834 00:52:46,000 --> 00:52:48,960 And she's kind of have to be like, no, I didn't mean in this 835 00:52:48,960 --> 00:52:51,720 way or like, I mean in this way. And like, no, like this is what 836 00:52:51,720 --> 00:52:52,800 I mean. Like instead of actually just 837 00:52:52,800 --> 00:52:55,360 being able to, to do it and just having the control. 838 00:52:55,360 --> 00:53:01,400 And so while someone who has never created some form of a 839 00:53:01,400 --> 00:53:04,200 digital artifact, we can now do that. 840 00:53:04,200 --> 00:53:08,160 And that's really cool. It is also, as you said, like 841 00:53:08,160 --> 00:53:09,920 it's really important to understand like what, what are 842 00:53:09,920 --> 00:53:16,280 we really risking if we throw away the the people who do it 843 00:53:16,280 --> 00:53:18,880 really well? And yeah, I think we risk a lot 844 00:53:18,880 --> 00:53:21,600 of lot of that collective collective gain or collective 845 00:53:21,600 --> 00:53:27,000 good. OK, so myth #2 Jenny, I doesn't 846 00:53:27,000 --> 00:53:32,800 boost creativity overall. Yeah, I would say that let me 847 00:53:32,800 --> 00:53:38,040 share one pro tip, a personal Aleen pro tip and I think this 848 00:53:38,760 --> 00:53:44,800 so this is related to using ChatGPT for for creative tasks. 849 00:53:45,720 --> 00:53:49,280 Some advice that I hear out there is, oh, use it to 850 00:53:49,280 --> 00:53:54,040 brainstorm and get ideas and like, like basically use it for 851 00:53:54,040 --> 00:53:57,640 your first draft of something. And I think that is a terrible, 852 00:53:57,640 --> 00:54:02,320 terrible, terrible idea. I think you absolutely have to 853 00:54:02,320 --> 00:54:05,760 use for all people use your brain as the first draft. 854 00:54:05,840 --> 00:54:09,560 If you need help structuring it, clarifying your ideas, seeing 855 00:54:09,560 --> 00:54:15,040 what you've missed, fine. But to, to use ChatGPT as your 856 00:54:15,040 --> 00:54:17,760 first draft, then, then what do you have? 857 00:54:17,760 --> 00:54:22,480 You now have a new status quo, which is this fully fleshed out 858 00:54:22,480 --> 00:54:25,240 report. And one thing we know as 859 00:54:25,240 --> 00:54:29,960 behavioral scientists about humans is we we are attached to 860 00:54:29,960 --> 00:54:32,280 the things that we own. So it's like this is the the 861 00:54:32,320 --> 00:54:35,040 endowment effect in in play right here. 862 00:54:35,040 --> 00:54:39,120 We don't want to lose by editing, getting rid of the 863 00:54:39,120 --> 00:54:42,520 suggestions that ChatGPT made. We don't want to lose that 864 00:54:42,520 --> 00:54:44,520 content. Yeah, you're, it's a great, 865 00:54:44,640 --> 00:54:47,200 great, great, great point. I think you as you're like 866 00:54:47,200 --> 00:54:52,760 alluding to like you basically sell yourself short by just 867 00:54:52,760 --> 00:54:57,640 molding yourself by the most generic responses given to you 868 00:54:57,640 --> 00:55:00,560 by Shajibati. It's really both important not 869 00:55:00,560 --> 00:55:04,360 to let these models and so on think for us too much like it's 870 00:55:04,360 --> 00:55:07,960 really useful to get feedback and, and and expand our thinking 871 00:55:07,960 --> 00:55:10,960 and test our thinking about not to do anything. 872 00:55:11,280 --> 00:55:16,440 And, and sadly, because it is actually what I've noticed 873 00:55:16,440 --> 00:55:20,760 myself, because I really agree with your advice, but at some 874 00:55:20,760 --> 00:55:26,080 time also noticing that it is tempting to start also being 875 00:55:26,080 --> 00:55:29,080 like, hey, give me 4 ideas, give me 4 things. 876 00:55:29,080 --> 00:55:32,880 Give me, give me. But instead of jumping that 877 00:55:32,880 --> 00:55:37,160 straight away, taking a little bit of a moment to reflect of 878 00:55:37,560 --> 00:55:40,640 again, what do I think, what are my thoughts, what I believe, 879 00:55:40,640 --> 00:55:43,720 what are my ideas? And then starting from there. 880 00:55:43,720 --> 00:55:45,520 And then you can always like build on that. 881 00:55:46,040 --> 00:55:49,800 I've never seen that not being the best way for like, I've 882 00:55:49,800 --> 00:55:52,560 never seen the way where I've started what he is like asking 883 00:55:52,560 --> 00:55:55,800 JPT to just do something for me and not in the end I feel I've 884 00:55:55,800 --> 00:55:59,200 wasted 30 minutes where it just giving me some generic stuff 885 00:55:59,200 --> 00:56:01,800 where I have to. In the end you start over and, 886 00:56:01,960 --> 00:56:05,120 you know, give it what I actually need. 887 00:56:05,120 --> 00:56:09,200 All right, so that was myth #2 Gen. 888 00:56:09,240 --> 00:56:11,880 AI boost creativity. No, it doesn't. 889 00:56:12,360 --> 00:56:14,920 Myth busted. OK, next one. 890 00:56:15,840 --> 00:56:22,840 So I think there's this this tendency to talk about AI as 891 00:56:22,840 --> 00:56:26,480 being this huge risk in terms of user privacy. 892 00:56:27,640 --> 00:56:32,480 And I think that this misses some really important nuance. 893 00:56:32,800 --> 00:56:38,160 And that is just that not all AI is equally unsafe. 894 00:56:38,160 --> 00:56:41,840 It's not all equally non private. 895 00:56:42,320 --> 00:56:46,440 And, and, and really you can create models depending you 896 00:56:46,440 --> 00:56:50,960 being a company, a product manager, whoever's in control of 897 00:56:51,200 --> 00:56:56,120 the AI product, you can, you can create it differently depending 898 00:56:56,120 --> 00:57:00,280 on your privacy goals. And some companies have really 899 00:57:00,320 --> 00:57:03,840 made a real effort towards towards using and developing 900 00:57:03,840 --> 00:57:08,480 privacy preserving AI. And this is a has become a whole 901 00:57:08,480 --> 00:57:11,800 subfield in itself. And you can look at what does 902 00:57:11,800 --> 00:57:14,880 that mean for maintaining privacy in your training data, 903 00:57:15,400 --> 00:57:17,600 in your input data and your output data. 904 00:57:18,000 --> 00:57:22,840 And one pretty popular way of doing this is called Federated 905 00:57:22,840 --> 00:57:25,640 learning. And this is basically just a 906 00:57:25,640 --> 00:57:27,520 version of on device machine learning. 907 00:57:27,520 --> 00:57:30,360 So the data never leaves the user's device. 908 00:57:30,720 --> 00:57:33,480 It means there are some implications of this, some 909 00:57:33,480 --> 00:57:36,320 downsides. You might say the the AI models 910 00:57:36,320 --> 00:57:39,720 have to be smaller. You can't fit ChatGPT on your 911 00:57:39,720 --> 00:57:45,480 phone. ChatGPT is extremely energy 912 00:57:45,480 --> 00:57:47,320 inefficient, so you can't do that. 913 00:57:48,080 --> 00:57:52,240 But your, your data never leaves your phone. 914 00:57:52,400 --> 00:57:55,120 Like that's a beautiful, important thing. 915 00:57:55,120 --> 00:57:59,680 And, and I feel like that is not talked about maybe as as much as 916 00:57:59,680 --> 00:58:02,040 it should be. So your, your personal data is 917 00:58:02,040 --> 00:58:03,160 totally secure. It is. 918 00:58:03,520 --> 00:58:07,040 It belongs to you and it isn't given to anyone else. 919 00:58:07,160 --> 00:58:11,360 Yeah, I think that also looking at comparing the large language 920 00:58:11,360 --> 00:58:15,160 models that are open source that you can kind of self host and 921 00:58:15,160 --> 00:58:18,440 and store locally compared to the ones that you can't like in 922 00:58:18,440 --> 00:58:20,920 the case of opening eyes or and tropics. 923 00:58:21,960 --> 00:58:25,240 The the difference is becoming I think less in many ways. 924 00:58:25,280 --> 00:58:29,880 And yeah, it's also both for the bigger models, if you wanted to 925 00:58:29,880 --> 00:58:34,480 use them with your company data and so on, they can certainly be 926 00:58:35,080 --> 00:58:39,560 self hosted and privately stored and nothing leaves the company. 927 00:58:39,880 --> 00:58:42,240 But as you said, like also what what is exciting is this kind of 928 00:58:42,240 --> 00:58:47,720 like increase kind of promise of these maybe tiny models that can 929 00:58:47,720 --> 00:58:50,840 exist even on the device. When I think that's kind of 930 00:58:50,840 --> 00:58:54,000 where, yeah, you, you obviously you can't maybe do exactly all 931 00:58:54,000 --> 00:58:57,560 of the things in the same way. But as you say, like it's you 932 00:58:57,560 --> 00:59:00,320 can do a lot of things, yeah. A lot of things, in fact. 933 00:59:00,520 --> 00:59:05,000 That leads me to our next myth, which is that more complex 934 00:59:05,000 --> 00:59:08,360 models are better because they often are not. 935 00:59:08,840 --> 00:59:11,600 There are quite a few downsides to having more complex models, 936 00:59:12,160 --> 00:59:14,160 including their lack of transparency. 937 00:59:14,160 --> 00:59:18,240 So when we talk about the black box, the bigger and more complex 938 00:59:18,240 --> 00:59:21,360 your model is, the less explainable it is. 939 00:59:21,680 --> 00:59:26,560 So you don't have what we call interpretable or explainable AI 940 00:59:26,560 --> 00:59:28,760 where you can say here's what the model is doing. 941 00:59:28,880 --> 00:59:32,360 So if, if you have a really big complex model, you can no longer 942 00:59:32,360 --> 00:59:34,520 do that. And that's that there's a very 943 00:59:34,520 --> 00:59:38,480 clear downside to that. If you don't understand it, you, 944 00:59:38,560 --> 00:59:41,720 you, you not only cannot tell your users what's going on, but 945 00:59:41,720 --> 00:59:44,560 you yourself don't know what's going on behind the scenes. 946 00:59:44,840 --> 00:59:47,120 And so that there's some obvious, some very real and 947 00:59:47,120 --> 00:59:51,120 obvious dangers there, but also the, the plastic downside of 948 00:59:51,120 --> 00:59:55,440 complex models is that you have a risk of overfitting. 949 00:59:55,440 --> 01:00:00,800 So this, this is what happens when you say you, you train 950 01:00:01,040 --> 01:00:05,520 some, a model on some piece of data, but you train it so 951 01:00:05,520 --> 01:00:11,080 closely to that data that, that it, it can only predict that 952 01:00:11,080 --> 01:00:12,960 data. When you then send it out into 953 01:00:12,960 --> 01:00:15,360 the real world, it does a, a pretty poor job. 954 01:00:15,360 --> 01:00:17,400 And this is overfitting to the training data. 955 01:00:18,440 --> 01:00:21,600 And then of course, more, bigger, more complex models are 956 01:00:21,600 --> 01:00:25,880 slower, they take more energy. So if you don't have to do that, 957 01:00:25,880 --> 01:00:30,560 if you can use a some a linear regression model, then do that 958 01:00:30,560 --> 01:00:33,600 by all means. Yeah, yeah, 100%. 959 01:00:34,040 --> 01:00:37,080 I think that's so true. And yeah, we covered quite a bit 960 01:00:37,080 --> 01:00:41,520 of myths there. Any any final myth or a little 961 01:00:41,520 --> 01:00:45,040 thing that you wanted to mention before we get to our final 962 01:00:46,480 --> 01:00:50,360 showdown of this episode? I think I would only say one 963 01:00:50,360 --> 01:00:55,520 final thing, and this is going to have to be an episode on its 964 01:00:55,520 --> 01:00:58,040 own because it's such a huge, huge topic. 965 01:00:58,080 --> 01:01:01,520 And this is one that Sendhil Mulanathan has, has really 966 01:01:01,520 --> 01:01:03,600 tackled. And this is the idea of 967 01:01:03,600 --> 01:01:06,840 algorithmic bias and human bias and, and sort of comparing the 968 01:01:06,840 --> 01:01:09,080 two. So the myth is that algorithmic 969 01:01:09,240 --> 01:01:12,840 bias is worse than human bias because in in the public 970 01:01:12,840 --> 01:01:16,760 conversation, we talk all about how terrible biases in our 971 01:01:16,760 --> 01:01:19,400 algorithms and how AI is so biased and so on. 972 01:01:19,680 --> 01:01:26,440 And Sendal makes the very good point that humans are very, very 973 01:01:26,440 --> 01:01:30,680 hard to change. Algorithms may be hard too, but 974 01:01:30,800 --> 01:01:33,600 surely not as hard as changing humans. 975 01:01:33,960 --> 01:01:38,720 And so that that that has been something that I think about a 976 01:01:38,720 --> 01:01:41,440 lot. And obviously work has to go 977 01:01:41,440 --> 01:01:46,520 into making algorithms non biased or biased in the in the 978 01:01:46,520 --> 01:01:49,760 correct way, right? Maybe correcting for bias out in 979 01:01:49,760 --> 01:01:55,280 the real world, but in in terms of the lift compared to changing 980 01:01:55,280 --> 01:01:59,480 humans, as behavioral scientists we know just how difficult it is 981 01:01:59,480 --> 01:02:01,880 to change people. So if we can just change some 982 01:02:01,880 --> 01:02:03,680 math. Not to understate the 983 01:02:03,680 --> 01:02:06,760 difficulty, but surely relatively much easier. 984 01:02:08,440 --> 01:02:10,400 Yeah. And yeah, Sandhu, if you're 985 01:02:10,400 --> 01:02:13,880 listening, let us know and and we'll book you in for episode. 986 01:02:14,920 --> 01:02:18,480 Awesome. So finally, we have a new quick 987 01:02:18,480 --> 01:02:22,000 fire round for this season. It's called to AI or not to AI. 988 01:02:22,440 --> 01:02:26,720 Basically, we're really interested to hear from my guest 989 01:02:26,720 --> 01:02:32,920 in terms of their thoughts about really using AI or not in the 990 01:02:32,920 --> 01:02:35,760 day-to-day or their work. And so we're kind of providing 991 01:02:35,760 --> 01:02:40,720 with some tasks and asking them, is this something that you think 992 01:02:41,200 --> 01:02:44,760 AI should do that it's well suited to or not? 993 01:02:45,200 --> 01:02:50,120 And so for this kind of first episode, we thought it'd be fun 994 01:02:50,120 --> 01:02:51,400 to use. Yeah. 995 01:02:51,840 --> 01:02:55,800 Ask each other a few of those to warm, warm ourselves up. 996 01:02:55,800 --> 01:02:59,920 And also just, yeah, we've kept this secret so we don't know 997 01:03:00,200 --> 01:03:03,880 what the other person can ask. But I have a few ready for you. 998 01:03:04,040 --> 01:03:05,480 Eileen, I'm. Ready. 999 01:03:05,920 --> 01:03:15,120 OK so my first one for you to AI or not to AI let your Tesla 1000 01:03:15,480 --> 01:03:19,720 drive your son to daycare on its own using autonomous mode. 1001 01:03:21,920 --> 01:03:25,160 Absolutely not. Never, never. 1002 01:03:25,320 --> 01:03:30,280 Or like another. Really hard to imagine a time 1003 01:03:30,280 --> 01:03:33,000 when I would be, Yeah. I mean, he certainly won't be in 1004 01:03:33,000 --> 01:03:35,920 daycare, but he probably won't even be in school by the time 1005 01:03:35,920 --> 01:03:38,920 that that autonomous vehicles are ready for that. 1006 01:03:39,440 --> 01:03:43,280 Even if you knew that you were probably on average a worse 1007 01:03:43,280 --> 01:03:45,640 driver than most autonomous vehicles, yeah. 1008 01:03:45,840 --> 01:03:48,120 Would you still? Would you still wanna do it 1009 01:03:48,120 --> 01:03:51,600 yourself? So I probably am a worse driver. 1010 01:03:53,040 --> 01:03:56,560 That's the reality, especially in the Tesla's too big for me. 1011 01:03:57,160 --> 01:04:02,040 But yeah, I don't trust it. I've had too many incidents 1012 01:04:02,040 --> 01:04:06,240 where it does a crazy thing and you have to you have to take 1013 01:04:06,240 --> 01:04:09,920 over. And it doesn't feel like the 1014 01:04:09,920 --> 01:04:13,400 technology is moving fast enough in order for me to trust it. 1015 01:04:13,400 --> 01:04:17,360 I want so badly for that day to come. 1016 01:04:17,480 --> 01:04:19,840 It just doesn't feel close at all. 1017 01:04:20,240 --> 01:04:23,080 So maybe my answer is eventually, but not before. 1018 01:04:23,240 --> 01:04:34,920 My son is an old man himself. OK Sam, for a podcast on 1019 01:04:34,920 --> 01:04:39,240 behavioral design, generate guest BIOS and episode summaries 1020 01:04:39,240 --> 01:04:42,440 and then combine that into a script for hosts to read 1021 01:04:45,920 --> 01:04:52,080 hypothetically. Hypothetically, some of it, yes, 1022 01:04:52,280 --> 01:04:53,640 some of it, no. Maybe. 1023 01:04:54,600 --> 01:04:57,720 I do think there's amazing things you can do today in terms 1024 01:04:57,720 --> 01:05:01,560 of if you have a very strong and clear understanding what the 1025 01:05:01,560 --> 01:05:05,320 output you want to be like, let's say like a bio or a 1026 01:05:06,880 --> 01:05:09,080 transcript or various things that you want to include in your 1027 01:05:09,080 --> 01:05:11,080 podcast. Definitely that can be done. 1028 01:05:11,160 --> 01:05:14,280 And maybe it's being done for for this podcast. 1029 01:05:14,400 --> 01:05:20,280 But in terms of like providing the extension of a longer script 1030 01:05:20,280 --> 01:05:25,440 or like actually having some form of artificial holds that 1031 01:05:25,440 --> 01:05:30,200 would do some form of thing, No, not not that, but but yeah, 1032 01:05:30,200 --> 01:05:33,240 introducing the guests, yeah, that kind of stuff, yeah. 1033 01:05:34,000 --> 01:05:38,680 Interesting. OK, OK, my turn. 1034 01:05:38,800 --> 01:05:43,160 OK, Eileen, to AI or not to AI? Create a virtual version of 1035 01:05:43,160 --> 01:05:47,480 yourself that speaks Swiss German really well to provide 1036 01:05:47,480 --> 01:05:51,240 care and companionship for your older relative. 1037 01:05:52,480 --> 01:06:01,840 Aw, that's would be gosh, I have such mixed feelings about that 1038 01:06:01,840 --> 01:06:03,240 because. They can chat with someone that 1039 01:06:03,480 --> 01:06:05,400 can maybe. Well, yeah, it's it's so 1040 01:06:05,400 --> 01:06:08,200 interesting. In the general case, I feel 1041 01:06:08,200 --> 01:06:14,360 pretty strong anti anti feelings to not AI. 1042 01:06:14,600 --> 01:06:19,280 However, in this specific case, Mike Grossi is 90 something 1043 01:06:19,280 --> 01:06:22,440 years old and it's not like I'm in Switzerland with her. 1044 01:06:22,800 --> 01:06:27,040 So it's not replacing me. I'm not there. 1045 01:06:27,960 --> 01:06:31,920 And so, gosh, wow, you, you really made made me see the flip 1046 01:06:31,920 --> 01:06:38,520 side that I hadn't seen before. OK, I'm going to say I'm willing 1047 01:06:38,520 --> 01:06:42,560 to try it, but now I can't let any of my relatives listen to 1048 01:06:42,560 --> 01:06:48,440 this episode. I'd like to oversee it for a few 1049 01:06:48,440 --> 01:06:49,520 years at least. OK. 1050 01:06:53,920 --> 01:07:00,120 OK. To AI or not to AIA marriage 1051 01:07:00,120 --> 01:07:06,120 counselor? Actually, it's funny because 1052 01:07:06,960 --> 01:07:12,000 that is my next to you. I have to not to AI have AI 1053 01:07:12,000 --> 01:07:14,400 relationship coach analyze your marriage and suggest 1054 01:07:14,400 --> 01:07:20,080 improvements. Yeah, I would say to AI 1055 01:07:20,080 --> 01:07:26,920 actually, I would be quite open to to getting feedback and yeah, 1056 01:07:27,600 --> 01:07:33,600 ideas from from an artificial marriage coach I think. 1057 01:07:33,640 --> 01:07:38,840 I think so too. I think so too because I think 1058 01:07:38,840 --> 01:07:47,720 it's easy for you as a couple to discredit the AI suggestions if 1059 01:07:47,720 --> 01:07:51,560 you don't like them. Easier than a person, and maybe 1060 01:07:51,560 --> 01:07:53,440 it'll bring you together to have a common enemy. 1061 01:07:54,600 --> 01:07:56,680 Yeah. And it's it's probably also 1062 01:07:56,680 --> 01:07:59,600 where it's one of the cases where at least initially short 1063 01:07:59,600 --> 01:08:03,520 term, it would be easier to probably be able to talk about 1064 01:08:03,520 --> 01:08:05,080 anything and be more open. As long as you feel like your 1065 01:08:05,080 --> 01:08:11,160 privacy is protected, then yeah, I wouldn't mind trying it in 1066 01:08:11,160 --> 01:08:14,720 terms of, and I would say this is a good example where I do 1067 01:08:14,720 --> 01:08:18,439 think long term, even though it could perform as well in terms 1068 01:08:18,439 --> 01:08:21,359 of use, like task for task perform as well. 1069 01:08:21,960 --> 01:08:24,680 There is I think some inherent value in actually having a human 1070 01:08:24,680 --> 01:08:28,760 being that you feel accountable to and that you trust and so on. 1071 01:08:28,800 --> 01:08:33,160 That's kind of there to support your relationship that will make 1072 01:08:33,160 --> 01:08:37,439 you work a little bit harder, be a little more engaged and so on. 1073 01:08:37,560 --> 01:08:41,399 Whereas you might not be as motivated to show up in time or 1074 01:08:41,399 --> 01:08:44,520 like do whatever thing you need to do and you are starting less 1075 01:08:44,520 --> 01:08:47,200 seriously than if you actually have to shop for a person and 1076 01:08:47,200 --> 01:08:50,600 pay for it and so on. So I think there is some part of 1077 01:08:50,600 --> 01:08:54,080 the traditional setup that I do think it's hard to replace in 1078 01:08:54,080 --> 01:08:57,200 that sense. So this one is a little bit 1079 01:08:57,399 --> 01:09:02,720 longer, but again, to AI or not to AI to use some form of AI 1080 01:09:02,720 --> 01:09:06,720 that tracks your day through, let's say your smartwatch or 1081 01:09:06,720 --> 01:09:08,560 smartphone or maybe smart glasses. 1082 01:09:09,160 --> 01:09:12,600 So it tracks your day. And then at the end of the day, 1083 01:09:12,600 --> 01:09:16,319 each night, it gives you feedback on whether you lived up 1084 01:09:16,960 --> 01:09:20,600 to your best self, like whether you like, you can set the 1085 01:09:20,600 --> 01:09:23,160 standard in some ways, like this is who I want to be. 1086 01:09:23,600 --> 01:09:27,160 And then every evening getting a little bit of like a mean a 1087 01:09:27,160 --> 01:09:32,840 report of like, hey, we'll talk. Eileen, here's how you've been 1088 01:09:33,279 --> 01:09:36,439 performing today. I think if I would be able to 1089 01:09:36,840 --> 01:09:42,640 set the seriousness of my commitment to that goal and also 1090 01:09:42,640 --> 01:09:47,640 not get that report on a daily basis, I think maybe less 1091 01:09:47,640 --> 01:09:50,080 frequent. I could handle like a weekly 1092 01:09:50,080 --> 01:09:54,400 check in or even a monthly check in and then at those check 1093 01:09:54,400 --> 01:09:59,280 insurance, set new goals for the next the next time period. 1094 01:09:59,280 --> 01:10:03,200 Yeah, I'd be up for that. I mean, if I'm controlling 1095 01:10:03,360 --> 01:10:05,920 everything, sure. I don't, yeah. 1096 01:10:06,760 --> 01:10:09,880 And and it's, I think I just say, I think it, I would 1097 01:10:09,880 --> 01:10:12,400 definitely be more up to it also like on a monthly basis, but 1098 01:10:12,760 --> 01:10:15,680 like on a daily basis, like I think eventually, because it's 1099 01:10:15,680 --> 01:10:20,040 very uncomfortable to be putting that pressure on yourself. 1100 01:10:20,040 --> 01:10:21,920 And I think that's what we're doing as society today. 1101 01:10:21,920 --> 01:10:24,560 And that's why I put it like it's more extreme to that. 1102 01:10:24,640 --> 01:10:27,040 But yeah, I'm sorry. What's your What's your final? 1103 01:10:28,080 --> 01:10:35,320 My final one it it's related to my your your Swiss question, but 1104 01:10:35,320 --> 01:10:41,240 this is in the afterlife. So to AI or not to AI your great 1105 01:10:41,400 --> 01:10:47,000 grandfather. So for example, in a virtual 1106 01:10:47,000 --> 01:10:53,640 reality world, you have a fully formed great grandfather who 1107 01:10:54,000 --> 01:10:57,560 walks like him, talks like him, you can talk to him, you can 1108 01:10:57,840 --> 01:11:01,240 play catch with him, you can visit him anytime you want. 1109 01:11:02,520 --> 01:11:04,600 Not really. No. 1110 01:11:06,560 --> 01:11:08,840 Yeah, yeah, it does Like this so much. 1111 01:11:08,840 --> 01:11:11,360 Speaks to all of my sci-fi interests in terms of like 1112 01:11:11,360 --> 01:11:15,000 Westworld and and there's an episode of Black Mirror that's 1113 01:11:15,000 --> 01:11:18,960 kind of mirrors this concept and. 1114 01:11:18,960 --> 01:11:25,440 But no, not for, not for me. I do think it's nice to be able 1115 01:11:25,440 --> 01:11:30,240 to let go and to, yeah, to. Maybe not nice, but important. 1116 01:11:30,400 --> 01:11:32,120 Important, Yeah. It's important to to be able to 1117 01:11:32,120 --> 01:11:34,200 let go. And it's important to in some 1118 01:11:34,200 --> 01:11:36,840 way, I feel like it also diminish what actually was there 1119 01:11:36,840 --> 01:11:39,200 before because you can like holding on to some form of 1120 01:11:39,200 --> 01:11:42,840 fragment of who they were or who they are, but it's actually not. 1121 01:11:43,520 --> 01:11:45,640 And it's not actually them. Yeah. 1122 01:11:46,840 --> 01:11:50,120 Yeah, this is this. This reminds me of since I'm on 1123 01:11:50,160 --> 01:11:56,720 a Sherry Turkle kick. She tells a story about meeting 1124 01:11:56,800 --> 01:12:01,520 a virtual Steve Jobs and she's done. 1125 01:12:01,560 --> 01:12:05,840 She's done some more recent work on artificial intimacy and 1126 01:12:06,200 --> 01:12:10,960 looking at people, for example, who have created a version of 1127 01:12:10,960 --> 01:12:15,080 someone that they knew who is who is passed on and they want 1128 01:12:15,080 --> 01:12:16,320 to continue interacting with them. 1129 01:12:16,320 --> 01:12:20,360 So she met a virtual Steve Jobs as part of this research and she 1130 01:12:20,360 --> 01:12:22,680 actually knew Steve Jobs when he was alive. 1131 01:12:23,480 --> 01:12:29,240 And her experience with the virtual Steve Jobs was, oh, like 1132 01:12:29,240 --> 01:12:32,160 maybe they uploaded his mannerisms and everything that 1133 01:12:32,160 --> 01:12:34,640 he's ever said. And they have this, the wealth 1134 01:12:34,760 --> 01:12:38,080 of everything on Steve Jobs. And he's a public figure. 1135 01:12:38,080 --> 01:12:41,520 So there's a whole lot of information about Steve Jobs out 1136 01:12:41,520 --> 01:12:44,200 there. But her experience was like, he 1137 01:12:44,200 --> 01:12:48,040 would never say that this is the, like, there are some things 1138 01:12:48,040 --> 01:12:53,200 that maybe feel like a likeness, but he would hate the virtual 1139 01:12:53,200 --> 01:12:55,960 Steve Jobs. Like, he would not approve. 1140 01:12:55,960 --> 01:13:00,240 Yeah. Fidelity, it's hard to, yeah, 1141 01:13:01,360 --> 01:13:04,280 establish. And yeah, I do think it's 1142 01:13:04,280 --> 01:13:07,160 interesting that we're both kind of shows a lot of these 1143 01:13:07,160 --> 01:13:10,560 questions that centered around significant relationships in our 1144 01:13:10,560 --> 01:13:14,120 life, like our connection to our kids or elder relatives or 1145 01:13:14,120 --> 01:13:16,800 partners, and I guess ourselves as well. 1146 01:13:17,160 --> 01:13:21,400 And I do think there's, I'm excited for this season to talk 1147 01:13:21,400 --> 01:13:27,160 about how AI can be in terms of either a tool or convenience 1148 01:13:27,160 --> 01:13:32,280 when it comes to recommending certain things regarding, I 1149 01:13:32,280 --> 01:13:36,560 guess classic stuff like recommended mention components 1150 01:13:36,560 --> 01:13:38,080 or recommending TV shows and so on. 1151 01:13:38,080 --> 01:13:41,000 I think that is interesting. But I do also think it's, it's 1152 01:13:41,000 --> 01:13:44,240 super interesting and profound to think about how AI can 1153 01:13:44,240 --> 01:13:46,760 influence who we are as, as people. 1154 01:13:46,760 --> 01:13:51,480 And kind of it does confront this notion of who are we as 1155 01:13:51,480 --> 01:13:53,400 humans? What are we? 1156 01:13:53,920 --> 01:13:57,520 What is it to be us? What is it not to be? 1157 01:13:57,520 --> 01:14:00,240 What is it be artificial? And, and how do we grapple with 1158 01:14:00,240 --> 01:14:02,800 that? And I think in some ways I think 1159 01:14:02,800 --> 01:14:06,320 that's, that's really real, especially as we starting to use 1160 01:14:06,320 --> 01:14:13,360 AI to be redefined as we have replica replicating partners, we 1161 01:14:13,360 --> 01:14:19,160 have a replicating, you know, customer service agents, these 1162 01:14:19,200 --> 01:14:23,360 things that is trying to imitate and as you say like creative 1163 01:14:23,360 --> 01:14:26,840 jobs or whatever. So it does gets to some very 1164 01:14:26,840 --> 01:14:30,800 deep and human questions and I I think if we can become a little 1165 01:14:30,800 --> 01:14:34,840 bit better understanding of how we can relate to those and how 1166 01:14:34,840 --> 01:14:37,480 we can kind of, yeah, feel comfortable there, I think 1167 01:14:37,480 --> 01:14:41,120 that's a big win as well in this season if we can approach on 1168 01:14:41,120 --> 01:14:43,480 this thing. So yeah, just a final reflection 1169 01:14:43,480 --> 01:14:44,880 on my end, but I think that's super interesting. 1170 01:14:45,440 --> 01:14:48,680 Yeah, I mean, all of these questions are just so perfectly 1171 01:14:48,680 --> 01:14:52,760 suited for behavioral science and just to toot our own horn 1172 01:14:52,760 --> 01:14:54,880 like that is why we have to do this. 1173 01:14:55,920 --> 01:15:00,360 I remember early on we were, we were sort of teeter tottering 1174 01:15:00,360 --> 01:15:04,120 about whether we really wanted to do AI because of the hype. 1175 01:15:05,200 --> 01:15:09,440 We felt the backlash and we decided that it would be 1176 01:15:09,440 --> 01:15:11,240 irresponsible of us not to do this. 1177 01:15:11,240 --> 01:15:14,240 So that's that's why we're here. Yeah. 1178 01:15:14,600 --> 01:15:17,200 And we're excited to share this with you and, and we in terms of 1179 01:15:17,200 --> 01:15:24,040 listeners and we have a e-mail address which is podcast at 1180 01:15:24,040 --> 01:15:27,360 habaweekly.com. And we're very, very keen to get 1181 01:15:27,360 --> 01:15:32,880 your thoughts, questions it asks people you want to hear from, 1182 01:15:33,320 --> 01:15:35,280 questions you want to get answered. 1183 01:15:36,000 --> 01:15:40,160 Send us all of that and anything or on that e-mail. 1184 01:15:40,160 --> 01:15:44,320 So podcast at habaweekly.com. And yeah, to wrap up this one, 1185 01:15:45,880 --> 01:15:48,320 get excited. We have the first interview 1186 01:15:48,920 --> 01:15:51,640 coming out next week. We have some fantastic, 1187 01:15:52,280 --> 01:15:54,560 honestly, I can say because we're recording them already, 1188 01:15:54,560 --> 01:15:57,200 some fantastic episodes that are already lined up for you. 1189 01:15:57,440 --> 01:16:02,560 So yeah, hopefully this was a little bit of a primer and yeah, 1190 01:16:02,920 --> 01:16:05,560 time to get excited for some some really interesting 1191 01:16:05,560 --> 01:16:09,920 conversations on paper science on AI and and beyond. 1192 01:16:09,920 --> 01:16:12,880 So so yeah, I'm excited, Eileen. I'm excited. 1193 01:16:13,480 --> 01:16:18,280 Yeah, and that's a wrap. You've been listening to the 1194 01:16:18,280 --> 01:16:21,640 Behavioral Design Podcast, brought to you by Habit Weekly 1195 01:16:21,640 --> 01:16:24,480 and Nuanced Behavior. Sam and Elaine tell me This 1196 01:16:24,480 --> 01:16:27,840 season is packed with incredible insights about behavioral design 1197 01:16:27,840 --> 01:16:31,480 and AI, so be sure to subscribe and share the podcast with your 1198 01:16:31,480 --> 01:16:33,240 friends. Though you might want to keep it 1199 01:16:33,240 --> 01:16:37,280 away from your enemies. In case you haven't noticed, I'm 1200 01:16:37,280 --> 01:16:40,680 an AI voice. Yep, pretty crazy. 1201 01:16:41,000 --> 01:16:43,960 Quite the improvement since last season's AI outro, don't you 1202 01:16:43,960 --> 01:16:46,680 think? If you'd like to collaborate 1203 01:16:46,680 --> 01:16:49,800 with us at Nuance Behavior, where we use behavioral design 1204 01:16:49,800 --> 01:16:52,600 to craft digital products with Nuance, e-mail us at 1205 01:16:52,600 --> 01:16:57,000 hello@nuancebehavior.com or book a call directly on our website, 1206 01:16:57,160 --> 01:17:01,680 nuancebehavior.com. A special thanks to the amazing 1207 01:17:01,680 --> 01:17:05,280 Dave Pizarro for our show music and to Mei Chen Yap and April 1208 01:17:05,280 --> 01:17:08,080 English for their help in producing and publishing this 1209 01:17:08,080 --> 01:17:10,320 episode. Thanks again for tuning in. 1210 01:17:10,600 --> 01:17:13,280 We'll be back soon with another exciting conversation where 1211 01:17:13,280 --> 01:17:15,840 behavioral design and AI intersect. 1212 01:17:17,080 --> 01:17:19,080 Happens. To. 1213 01:17:21,920 --> 01:17:56,360 Mugatroid. The. 1214 01:18:16,360 --> 01:18:20,200 The brain is beautiful and really, if you think machine 1215 01:18:20,200 --> 01:18:24,400 learning models can be a black box, tell me how consciousness 1216 01:18:24,400 --> 01:18:26,600 arises from our meat machines.