WEBVTT - AI & Us 0:00:04.600 --> 0:00:17.560 Sleepwalkers is a production of our Heart Radio and Unusual Productions. Hi, 0:00:17.720 --> 0:00:20.880 I'm Aloan and I'm care Price. Welcome to a special 0:00:20.960 --> 0:00:36.760 bonus episode of Sleepwalkers. Well, Cara, it's very good to 0:00:36.760 --> 0:00:39.000 be back to the office. That you It is good 0:00:39.000 --> 0:00:41.120 to be back, as I just want to apologize to 0:00:41.120 --> 0:00:43.720 our listeners. We don't have an algorithm that's going to 0:00:43.760 --> 0:00:48.320 create season two, so things have taken a little longer. Yes, unfortunately, 0:00:48.520 --> 0:00:51.680 unfortunately a I can't do everything yet. But we are 0:00:51.680 --> 0:00:54.000 hard at work on season two of Sleepwalkers, and we're 0:00:54.040 --> 0:00:57.600 focusing on stories that really contextualize the implications of AI, 0:00:57.960 --> 0:01:00.520 what it's doing, what the future is, and how it's 0:01:00.520 --> 0:01:03.400 affecting us. You know, we had such a good time 0:01:03.560 --> 0:01:06.560 in season one of Sleepwalkers wrapping our heads around the 0:01:06.560 --> 0:01:10.200 meaning of artificial intelligence. It becomes basically a principle of 0:01:10.240 --> 0:01:14.360 statistics prediction, how we're using data to inform our decisions, 0:01:14.560 --> 0:01:18.200 and how that's becoming ingrained in products and services and 0:01:18.400 --> 0:01:22.080 everything we do. Really, it's true the Pandora's box of 0:01:22.120 --> 0:01:24.520 AI has been opened, but we still have the black 0:01:24.560 --> 0:01:27.880 box problem that is true explainable AI. We can't tell 0:01:27.880 --> 0:01:30.360 what neural networks are doing yet, but people are working 0:01:30.400 --> 0:01:32.000 on it, and that's a story we're going to be 0:01:32.000 --> 0:01:35.440 covering closely in season two. But in this bonus episode 0:01:35.480 --> 0:01:37.440 of Sleepwalkers, we're going to take a look back at 0:01:37.440 --> 0:01:40.520 some of the most poignant stories and interesting applications of 0:01:40.520 --> 0:01:43.399 AI that we talked about in the first season, and 0:01:43.520 --> 0:01:45.360 later in this episode, I'm going to give you a 0:01:45.400 --> 0:01:48.280 preview of a fascinating new company that's using AI to 0:01:48.320 --> 0:01:54.200 predict very specific consumer taste, as in preferences, not like 0:01:54.320 --> 0:01:57.440 tasting clothing. Before we get through though us, when you 0:01:57.520 --> 0:01:59.640 look back at season one, what stands out to you 0:01:59.640 --> 0:02:01.560 the most? One of the things that stands out to 0:02:01.600 --> 0:02:04.320 me most is the story you did about liar Bird. 0:02:04.400 --> 0:02:07.880 Thank you, but seriously, the way they use an algorithm 0:02:07.960 --> 0:02:09.920 to create a deep fake of your voice. But in 0:02:09.960 --> 0:02:13.560 that particular piece, the questions it raised about mortality and 0:02:13.600 --> 0:02:16.079 would you want to hear your father's voice beyond the 0:02:16.160 --> 0:02:19.000 grave has stuck with me and I was one of 0:02:19.000 --> 0:02:21.639 the most powerful moments in the whole podcast. Yeah so, 0:02:22.000 --> 0:02:24.959 Jose Satello, one of the founders of Liar Bird, had 0:02:25.000 --> 0:02:27.080 me sit down on a microphone for an hour and 0:02:27.120 --> 0:02:30.800 just speak which was a personal dream of mine, and 0:02:30.840 --> 0:02:36.160 then using that out me interrupting. Using that voice data, 0:02:36.440 --> 0:02:39.400 liar Bird's algorithms created a version of my voice that 0:02:39.400 --> 0:02:42.200 could turn any written text into something that sounded like me. 0:02:42.720 --> 0:02:46.280 But Jose actually explained it better. So can you hit 0:02:46.320 --> 0:02:49.280 the clip? Not an AI scientist, but we do have 0:02:49.360 --> 0:02:53.600 the sophistication to roll tape. I know it might sound 0:02:53.639 --> 0:02:56.919 a bit like magic, but in reality, the way that 0:02:57.160 --> 0:02:59.919 our algorithm's work is basically they are just a battern 0:03:00.080 --> 0:03:04.680 matching algorithms, and so it's trying to figure out how 0:03:05.240 --> 0:03:09.160 to identify the patterns in your voice by comparing it 0:03:09.320 --> 0:03:12.800 against thousands of other voices a shoually tens of thousands 0:03:12.840 --> 0:03:16.000 of other voices, and trying to figure out what is 0:03:16.040 --> 0:03:19.480 it that makes your voice unique. Once Jose's algorithms identified 0:03:19.520 --> 0:03:22.919 what was unique about my voice, obviously everything they had 0:03:22.919 --> 0:03:25.639 the building blocks they needed to make a fake. Then 0:03:25.639 --> 0:03:28.120 we sent Jose a set of sentences we wanted robot 0:03:28.160 --> 0:03:31.040 care to say, and he used another set of algorithms 0:03:31.080 --> 0:03:33.359 to turn the text into what we heard. The way 0:03:33.400 --> 0:03:35.360 they do this is they use it's called a generative 0:03:35.480 --> 0:03:40.440 adversarial network again, which is a system where one neural 0:03:40.480 --> 0:03:43.720 net tries to trick another one a thousand times per second, 0:03:44.240 --> 0:03:47.120 So each time the second network to texts of fake, 0:03:47.560 --> 0:03:50.960 the first one tries again. It basically learns from its mistakes, 0:03:51.000 --> 0:03:53.800 and once it tricks its adversary, it's ready to show 0:03:53.800 --> 0:03:57.440 its results. In our case, liar bird pits my fake 0:03:57.520 --> 0:04:00.200 voice against my real voice until it sounds like this 0:04:00.800 --> 0:04:04.800 sub dog Scara Karen. One of the reasons why this 0:04:04.840 --> 0:04:07.160 story has stuck with me is because it feels like 0:04:07.200 --> 0:04:10.360 we're just at the beginning of tapping the potential and 0:04:10.400 --> 0:04:14.480 the potential for harm of deep fakes. May be remembered 0:04:14.480 --> 0:04:17.240 as the year of the first significant deep fake crime. 0:04:17.720 --> 0:04:20.920 An employee at a UK based energy firm believed he 0:04:21.000 --> 0:04:23.600 was on the phone to his boss and followed instructions 0:04:23.640 --> 0:04:27.560 to transfer two hundred thousand pounds to a scammers bank account. 0:04:28.160 --> 0:04:30.720 That certainly won't be the last deep fatefore we hear of, 0:04:30.960 --> 0:04:35.200 and it raises questions about responsibility and accountability. Who's liable 0:04:35.240 --> 0:04:38.000 in a case like this. Facebook has actually gone so 0:04:38.040 --> 0:04:41.320 far as to create a deep fake detection challenge to 0:04:41.400 --> 0:04:44.760 get the best minds thinking about deep fakes and how 0:04:44.760 --> 0:04:46.960 we might solve the problem and offering like a million 0:04:46.960 --> 0:04:50.200 dollar prize reach prize, which is like a dollar but 0:04:50.240 --> 0:04:52.800 it also shows how important the issue is, especially when 0:04:52.800 --> 0:04:55.279 a company like Facebook gets behind it. Um. There's another 0:04:55.320 --> 0:04:59.720 side to deep fake technology that actually highlights this dichotomy 0:04:59.760 --> 0:05:01.600 and tech anology right now, which is that it can 0:05:01.640 --> 0:05:05.120 be used for menacing purposes but also really powerful and 0:05:05.160 --> 0:05:08.560 beautiful applications. Jose goes on to talk about how liar 0:05:08.640 --> 0:05:11.000 Bird can be used with als patients and give them 0:05:11.040 --> 0:05:14.560 the ability to speak when they've lost all ability to speak, 0:05:14.800 --> 0:05:16.960 when it could give them the opportunity to speak in 0:05:17.120 --> 0:05:19.920 a version of their voice to their children again, which 0:05:19.960 --> 0:05:23.200 is quite profound. One area where I think technology is 0:05:23.200 --> 0:05:26.560 a powerhouse for change is in medicine. Technologists and doctors 0:05:26.560 --> 0:05:30.080 alike are looking at AI to predict, treat, and diagnose, 0:05:30.520 --> 0:05:33.359 you know, everything from depression to cancer, and that's a 0:05:33.480 --> 0:05:36.440 very wide spectrum. And it reminds me of one of 0:05:36.480 --> 0:05:39.880 my favorite interviews that you did, which was your interview 0:05:39.920 --> 0:05:43.400 with Saddartha. Muker g just so fascinated by this article 0:05:43.440 --> 0:05:46.000 he'd written for The New Yoker called AI versus m 0:05:46.080 --> 0:05:48.680 D which laid out all of the kind of benefits 0:05:48.680 --> 0:05:52.480 and potential applications of using AI in medicine, including some 0:05:52.640 --> 0:05:54.800 of the downside such as the black box problem of 0:05:54.839 --> 0:05:57.720 AI that you mentioned, not knowing why an algorithm has 0:05:57.720 --> 0:06:00.400 made a recommendation, and also another problem, which is that 0:06:00.440 --> 0:06:02.840 if we really too much on technology, it can erode 0:06:02.920 --> 0:06:05.920 human skills. There is a fear that AI could move 0:06:06.000 --> 0:06:08.039 us into a very black and white way of thinking. 0:06:08.400 --> 0:06:11.080 The computer says you have cancer, or the computer says 0:06:11.400 --> 0:06:14.040 you to have your liver removed, said Arthur, who is 0:06:14.080 --> 0:06:16.920 one of the world's foremost oncologists and a Pullet Surprise 0:06:16.960 --> 0:06:21.440 winning author. Provided a different perspective. There is something very 0:06:21.480 --> 0:06:25.880 fundamental about the human brain, a scientist's brain, a doctor's brain, 0:06:25.920 --> 0:06:29.640 and artist's brain that asks questions in a fundamentally different manner, 0:06:29.960 --> 0:06:32.400 the why question. Why did this happen in this person 0:06:32.480 --> 0:06:35.719 in this time? Why does the melanoma appear in the 0:06:35.720 --> 0:06:38.280 first calase? What is the molecular basis of that appearance. 0:06:38.480 --> 0:06:41.640 The most interesting mysteries of medicine remain mysteries that have 0:06:41.720 --> 0:06:45.159 to do with the why. Once we give up some 0:06:45.320 --> 0:06:49.440 of the diagnostic pattern recognition material to machines, it will 0:06:49.480 --> 0:06:50.919 be time to play. It will be the time to 0:06:50.960 --> 0:06:54.960 play in the arena of human therapeutics, human biology, the 0:06:55.040 --> 0:06:58.680 complexity of the human interaction, the art of medicine. My 0:06:58.800 --> 0:07:01.360 hope is that medicine, in being more playful, will become 0:07:01.760 --> 0:07:06.120 more compassionate, more able to take into account individuals and 0:07:06.160 --> 0:07:10.920 their individual destinies rather than bucketing people in big categories. 0:07:11.720 --> 0:07:15.360 It means having more time to spend with humans. You know, 0:07:15.400 --> 0:07:19.360 we are so constrained by time that even compassion gets 0:07:19.360 --> 0:07:23.520 three minutes, We won't become more robotic, will become less 0:07:23.600 --> 0:07:28.520 robotic as the robots enter our own So Dartha's point 0:07:28.680 --> 0:07:32.040 is that these tools could make doctors more efficient so 0:07:32.080 --> 0:07:34.440 that they can provide better care. It sort of takes 0:07:34.480 --> 0:07:39.720 the grunt work out of medicine and puts the patient 0:07:39.880 --> 0:07:43.120 care work back in the doctor's hands. This idea that 0:07:43.200 --> 0:07:46.600 technology can actually allow us to be more human, make 0:07:46.680 --> 0:07:50.120 us more empathetic, is fascinating, and it also raises the 0:07:50.200 --> 0:07:52.920 questions about new types of skills that may need to 0:07:52.920 --> 0:07:55.760 be developed in an age of AI. Yeah, and Regina 0:07:55.800 --> 0:07:58.360 Barslay from m I T spoke a lot about this. 0:07:58.560 --> 0:08:02.360 How doctors have to now equip themselves with new ways 0:08:02.520 --> 0:08:06.480 of translating data to patients, we still do not communicate 0:08:06.520 --> 0:08:10.040 it to the patient because I think now there is 0:08:10.080 --> 0:08:13.080 a walk to be done, not on computer science or AIPAD, 0:08:13.560 --> 0:08:16.200 but really on the clinical side. What is the best 0:08:16.240 --> 0:08:20.280 way to communicate it to the patient and what is 0:08:20.720 --> 0:08:23.920 um you know, the past that you're going to give them. 0:08:23.920 --> 0:08:25.920 It is not just enough to say you know you 0:08:25.960 --> 0:08:30.600 are high risk. You need to propose some suggestion and solution. 0:08:31.200 --> 0:08:35.520 So currently the clinical stuff is thinking and looking at 0:08:35.520 --> 0:08:39.760 the ways of effective you know, clinical engagement with a patient. 0:08:40.480 --> 0:08:42.880 You know, I speaking of data. You all know Harri 0:08:43.080 --> 0:08:45.520 was another person who made you and me think about 0:08:45.600 --> 0:08:49.840 humans as reducible to data. I think he's mostly known 0:08:49.960 --> 0:08:52.920 as a historian and for his book Sapiens, but you 0:08:52.960 --> 0:08:55.280 spoke with him about the data we produce as humans 0:08:55.320 --> 0:08:58.439 and how that influences our relationship with technology. That's right, 0:08:58.440 --> 0:09:00.760 which is the topic of his book Daus. And he 0:09:00.800 --> 0:09:03.760 has this phrase data is m to describe how we've 0:09:03.840 --> 0:09:07.200 kind of come to worship the data we create and 0:09:07.360 --> 0:09:12.400 our own technological creations. So what happens when based on 0:09:12.440 --> 0:09:15.520 all of our past behavior, AI starts to know us 0:09:15.640 --> 0:09:18.440 better than we know ourselves. Here's a clip from You 0:09:18.520 --> 0:09:22.320 are talking about exactly that. When we talk about AI, 0:09:22.600 --> 0:09:26.640 we tend to greatly exaggerate the potential abilities, but at 0:09:26.679 --> 0:09:30.560 the same time we also tend to exaggerate the abilities 0:09:30.640 --> 0:09:35.360 of humans. People say that AI is not going to 0:09:35.440 --> 0:09:39.080 take over our lives because it's very imperfect and it 0:09:39.160 --> 0:09:42.880 won't be able to know us perfectly. But what people 0:09:43.000 --> 0:09:47.280 forget is that humans often have a very poor understanding 0:09:47.600 --> 0:09:52.719 of themselves, of the desires, of their emotions, of their 0:09:52.760 --> 0:09:57.400 mental states. For AI to take over your life, it 0:09:57.559 --> 0:10:01.160 doesn't need to know you perfectly, just need to know 0:10:01.280 --> 0:10:04.720 you better than you know yourself. And that's not very 0:10:04.720 --> 0:10:08.360 difficult because we often don't know the most important things 0:10:08.480 --> 0:10:11.160 about ourselves. So, but let's say you could turn back 0:10:11.160 --> 0:10:14.680 the clock to being fifteen, would you have wanted to 0:10:14.800 --> 0:10:17.560 live in a world where there was sufficiently good sensors 0:10:17.640 --> 0:10:21.400 to monitor your eyes, your eye movement, you're breathing, you know, 0:10:21.400 --> 0:10:23.640 while you're going about your daily life, and then to 0:10:23.720 --> 0:10:27.199 interpret that and say to you you have all more 0:10:27.240 --> 0:10:30.080 likely than not you're gay. That's a very good question 0:10:30.440 --> 0:10:34.080 which will become very practical questions in a few years. 0:10:34.440 --> 0:10:38.240 And the way that I grew up and developed. It 0:10:38.320 --> 0:10:42.480 would have been a very bad idea. I wouldn't like 0:10:42.760 --> 0:10:46.480 to receive this kind of insight from form a machine. 0:10:46.800 --> 0:10:49.640 I'm not sure how I would have dealt with it 0:10:49.760 --> 0:10:51.760 when I was fifteen, you know, in Israel in the 0:10:51.840 --> 0:10:54.840 nineteen eighties, and maybe partly it was you know, a 0:10:54.880 --> 0:10:57.800 defense mechanism. In the future too. It it depends where 0:10:57.840 --> 0:11:02.400 you live. Brunei has instituted the death penalty for gay people, 0:11:02.440 --> 0:11:05.720 at least for people engaged in homosexual sex. So if 0:11:05.760 --> 0:11:08.480 I'm a teenager in Brunei, I don't want to be 0:11:08.520 --> 0:11:11.720 told by the computer that I'm gay, because the computer 0:11:11.800 --> 0:11:14.079 will then be able to tell that to the police 0:11:14.120 --> 0:11:17.679 and to the authorities as well. So the apps we use, 0:11:17.800 --> 0:11:20.320 the product we buy, the number of steps we take, 0:11:20.679 --> 0:11:23.920 the delivery I ordered last night, that will becomes data, 0:11:24.400 --> 0:11:27.080 and that data can feed into neural networks to create 0:11:27.120 --> 0:11:30.480 statistical models of us and what we might do next, 0:11:30.920 --> 0:11:33.920 sometimes in order to diagnose a medical condition, and other 0:11:33.960 --> 0:11:37.720 times to sell us a product. Here's V again, looking 0:11:37.760 --> 0:11:41.320 to the future, say ten twenty years. The danger is 0:11:41.600 --> 0:11:43.720 if I still don't know that I'm gay, but the 0:11:43.800 --> 0:11:47.920 government and Coca Cola and and Amazon and Google. They 0:11:47.920 --> 0:11:51.680 already know it. I'm at a huge disadvantage. So it 0:11:51.760 --> 0:11:55.600 could be something as frightening as the secret police coming 0:11:55.720 --> 0:11:58.880 and taking me to a concentration camp. But it could 0:11:58.960 --> 0:12:03.199 also be something Coca Cola knowing that I'm gay, they 0:12:03.200 --> 0:12:05.959 want to sell me a new drink, and they choose 0:12:06.040 --> 0:12:10.480 the advertisement with the shirtless guy and not the advertisement 0:12:10.840 --> 0:12:13.280 with the girl in the bikini. And next day morning 0:12:13.400 --> 0:12:15.880 I go and I buy this soft drink and I 0:12:15.920 --> 0:12:19.520 don't even know why, and they have this huge advantage 0:12:19.559 --> 0:12:23.480 over me and can manipulate me in all kinds of ways. Well, 0:12:23.480 --> 0:12:26.760 as you've all brought up soda, I was not allowed 0:12:26.760 --> 0:12:29.960 to drink soda as a child. My parents tricked me 0:12:30.000 --> 0:12:33.640 into thinking that Seltzer was soda. I later found out 0:12:33.720 --> 0:12:38.000 that soda is soda and Seltzer is water. And somehow 0:12:38.080 --> 0:12:41.640 the Seltzer of it all is the perfect segue because 0:12:41.679 --> 0:12:44.400 AI is not only being used to sell a product, 0:12:45.160 --> 0:12:49.240 it's also being used to create products like Seltzer in 0:12:49.280 --> 0:12:52.640 the R and D research and development phase. And for 0:12:52.679 --> 0:12:56.960 this bombs episode, Julian remember Juliana, a lovely producer, And 0:12:57.080 --> 0:13:00.400 I went to meet the company behind the gastrograph app, 0:13:00.440 --> 0:13:04.600 which is using consumer preference data to make predictions about 0:13:04.760 --> 0:13:07.920 new flavors. People might like, I think you hate Nettle, 0:13:08.640 --> 0:13:15.680 think against Analytical Flavor Systems or a f S is 0:13:15.720 --> 0:13:19.600 tucked away down a side street in Chinatown, up in 0:13:19.640 --> 0:13:23.280 a third story walk up. It does actually and when 0:13:23.280 --> 0:13:25.760 we were there, the office was still waiting for furniture. 0:13:25.800 --> 0:13:27.840 You know, it had this We're going to disrupt the 0:13:27.880 --> 0:13:31.360 industry vibe. We moved in like a month or two ago. 0:13:31.640 --> 0:13:34.760 But it turns out you can't just buy office furniture. 0:13:36.040 --> 0:13:38.720 We met the founder of the company, Jason Cohen, and 0:13:38.760 --> 0:13:41.079 we believe that in the future, in order to be competitive, 0:13:41.160 --> 0:13:43.680 you have to be targeted. That that there won't be 0:13:43.760 --> 0:13:47.360 any billion dollar brands right in ten years if you 0:13:47.400 --> 0:13:49.800 don't have an m b A. Here's what Jason's talking about. 0:13:50.040 --> 0:13:52.360 Think about the coffee you drank this morning, is it 0:13:52.559 --> 0:13:55.560 Third Wave? Did you get it from Starbucks or an 0:13:55.600 --> 0:13:59.040 indie roaster. Food and beverage companies are moving more and 0:13:59.120 --> 0:14:01.840 more towards knee markets. The problem is that they have 0:14:02.000 --> 0:14:05.120 very old school ways of developing new products. But a 0:14:05.280 --> 0:14:08.560 FS is offering another way to reach those customers, and 0:14:08.600 --> 0:14:13.080 that's by allowing companies to formulate more specific products using AI. 0:14:13.400 --> 0:14:16.079 Here's Jason again. Usually the way that things are done 0:14:16.120 --> 0:14:20.760 today is you get some conceptual brief. You might say, 0:14:20.800 --> 0:14:23.400 we want to develop a new fruit flavored beverage for 0:14:24.160 --> 0:14:27.360 Japanese millennial women. Right, you would look at what other 0:14:27.400 --> 0:14:29.840 fruit flavored beverages are out there. You'd look at your 0:14:29.840 --> 0:14:31.600 own product lineup and say, well, we already have a 0:14:31.680 --> 0:14:35.040 lemon flavor, and we're going to send out these briefs, 0:14:35.080 --> 0:14:36.880 and we're gonna send these out to these flavor companies. 0:14:36.880 --> 0:14:38.920 We're gonna see what other fruits we can get, and 0:14:38.920 --> 0:14:41.360 you're gonna wind up with very mainstream things. You're gonna 0:14:41.400 --> 0:14:45.040 wind up with peach and mango and strawberry and grapefruit, right, 0:14:45.040 --> 0:14:48.680 and maybe you'll wind up with something interesting like low 0:14:48.800 --> 0:14:53.360 quad or uzo or dragon fruit. Right. And then you're 0:14:53.360 --> 0:14:57.680 gonna have your own consumer tasting panel internally hopefully uh 0:14:57.680 --> 0:14:59.520 and you're gonna have them taste it, and they're gonna 0:14:59.560 --> 0:15:02.240 have to like some of those more than your current 0:15:02.280 --> 0:15:05.520 offering or more than a competitors. After you've done all 0:15:05.520 --> 0:15:07.360 of this work, which sometimes costs in the tens of 0:15:07.360 --> 0:15:09.680 thousands of dollars in order to have the samples developed. 0:15:09.680 --> 0:15:11.760 How the samples sent to you recruit the consumers, but 0:15:11.960 --> 0:15:14.320 the product in front of the consumers, that data is 0:15:14.400 --> 0:15:17.360 only ever usable once. Right. All you get from that 0:15:17.400 --> 0:15:20.360 as a binary yes no, sixty percent of the population 0:15:20.400 --> 0:15:23.280 liked more than the competitors, and so what we're doing 0:15:23.560 --> 0:15:27.960 is entirely different. Jason's team wants to take product development 0:15:28.000 --> 0:15:31.120 out of the yes no binary. Instead of just saying 0:15:31.320 --> 0:15:35.040 coke or pepsi, they can calculate which parts of each 0:15:35.080 --> 0:15:38.760 soft drink people liked and disliked, and then a f 0:15:38.920 --> 0:15:41.480 s can make entirely new flavors based on what a 0:15:41.520 --> 0:15:44.600 person kind of liked about pepsi and kind of liked 0:15:44.600 --> 0:15:48.280 about coke. And finally, they can transfer those preferences to 0:15:48.640 --> 0:15:52.600 entirely different demographics. So what might someone in Mexico want 0:15:52.640 --> 0:15:55.280 to taste in their cola compared to what a Japanese 0:15:55.280 --> 0:15:58.720 millennial woman might want to taste in her cola? This 0:15:58.800 --> 0:16:02.080 is what Jason believes is truly disruptive, being able to 0:16:02.080 --> 0:16:04.400 say to a brand, if you want to launch in Mexico, 0:16:04.480 --> 0:16:07.360 you should tweak your flavors in this way because we're 0:16:07.400 --> 0:16:10.000 actually able to collect this data, develop a data set, 0:16:10.240 --> 0:16:12.720 and make predictions. So we could take data from say, white, 0:16:12.720 --> 0:16:15.040 twenty to three year old college educated males, and use 0:16:15.080 --> 0:16:17.240 that to predict what every other population in the United 0:16:17.280 --> 0:16:19.120 States is going to perceive in that product. And so 0:16:19.160 --> 0:16:21.520 we're taking an industry that has never seen predictions of 0:16:21.520 --> 0:16:24.960 any kind before and finally being able to actually predict things, 0:16:25.120 --> 0:16:27.280 predict who's gonna like it, how much they're gonna like 0:16:27.360 --> 0:16:29.360 it right, and what we can do to optimize it 0:16:29.440 --> 0:16:31.600 so that they like it more. We can actually create 0:16:31.640 --> 0:16:33.440 products that no one would have ever thought of, and 0:16:33.560 --> 0:16:35.200 no one ever would have thought that a segment of 0:16:35.240 --> 0:16:37.160 the population would have liked. And this is something that 0:16:37.200 --> 0:16:38.920 we now do with the companies that we work with. 0:16:39.160 --> 0:16:41.760 We did talk quite a bit about developing a pine 0:16:41.800 --> 0:16:45.240 flavored beverage in Japan. When we first showed these results 0:16:45.240 --> 0:16:47.720 to a company there, they said, natzu desca, do you 0:16:47.760 --> 0:16:50.920 mean pineapple? Because it was just so out of the yeah, 0:16:50.960 --> 0:16:54.320 out of the ordinary. The way Jason and his team 0:16:54.360 --> 0:16:56.640 are able to get such nuanced data is with an 0:16:56.680 --> 0:16:59.960 app they developed called Gastrograph. Gastrograph looks a lot like 0:17:00.200 --> 0:17:02.800 the flavor wheels they use in coffee and wine tasting 0:17:02.840 --> 0:17:04.840 to help people map out what they taste when they 0:17:04.880 --> 0:17:07.520 try a new product. We think of every flavor, romance, 0:17:07.520 --> 0:17:10.000 texture as a signature. You can have the five basic 0:17:10.040 --> 0:17:13.240 flavors bitter, sweet, salty, sour mommy, and then underneath that 0:17:13.240 --> 0:17:16.080 you're gonna have categorical flavors like fruity earth, the herbaceous nuts, 0:17:16.080 --> 0:17:18.800 and seedge roasted. I mentioned that he's a professional taster, right, 0:17:18.960 --> 0:17:22.240 and then underneath that you can have subcategorical like citrus, 0:17:22.320 --> 0:17:24.840 or specific like lemon are very specific like Meyer lemon 0:17:24.840 --> 0:17:27.280 are very very specific like Meyer lemons as right, So 0:17:27.400 --> 0:17:31.199 all of those signatures exist in some infinite dimensional space 0:17:32.080 --> 0:17:36.480 flavor space. So, car, since you're such a Seltzer fiend, 0:17:36.560 --> 0:17:38.919 we demo gaster graph and got a feel for the 0:17:38.960 --> 0:17:42.040 app by doing a Seltzer tasting. Yeah, we tasted five 0:17:42.119 --> 0:17:44.560 Seltzers that are already on the market that you'd buy, 0:17:44.640 --> 0:17:47.920 like seven eleven, and we rated different components of them. 0:17:48.000 --> 0:17:50.359 So if I tasted fruit, I'd rate that from zero 0:17:50.440 --> 0:17:54.040 to five, and I could add adjectives like strawberry or mango. 0:17:54.320 --> 0:17:56.920 You know, I'm not into like Seltzer two point oh 0:17:56.920 --> 0:18:01.919 with like kumquat flavored sparkling seltzer. You know, I plane vintage, 0:18:02.400 --> 0:18:05.720 but a f s is Resident chemist Ryan On agreed 0:18:05.760 --> 0:18:09.399 to formulate a seltzer based on just our extremely small 0:18:09.480 --> 0:18:12.560 data set. Um, you'd probably be a pretty fast process. 0:18:12.600 --> 0:18:15.520 So we have tons of seltzer data. What we would 0:18:15.520 --> 0:18:19.000 want to do is um run through a couple of 0:18:19.000 --> 0:18:20.840 different flavors to get an idea of the types of 0:18:20.880 --> 0:18:23.920 things that you like, build a model specifically around that, 0:18:24.520 --> 0:18:27.280 run an optimization, predict a new flavor that you've never 0:18:27.320 --> 0:18:31.120 had before, and then have you tried again. So after 0:18:31.160 --> 0:18:34.760 we did all of that, we went home. We sat 0:18:34.800 --> 0:18:38.440 in our hands, and then we went back to the office, 0:18:38.600 --> 0:18:41.360 which actually a little bit more furniture when we got 0:18:41.400 --> 0:18:45.560 back there to see if they actually could create a 0:18:45.600 --> 0:18:48.359 seltzer that both Julian and I would like. So it 0:18:48.440 --> 0:18:51.200 was a blind tasting. Here we go, all right, drink 0:18:51.200 --> 0:18:56.399 one pair. I don't know why I think such a 0:18:56.400 --> 0:19:02.680 thing against pears. The number two berries, that's delish. Let's 0:19:02.680 --> 0:19:05.600 see what we got here. I wouldn't know huckleberry if 0:19:05.640 --> 0:19:10.119 it hit me in the face, honestly, but whatever, Number 0:19:10.119 --> 0:19:18.119 three honestly, curry. Do you not taste curry like there's spice? 0:19:18.200 --> 0:19:24.920 It's so interesting? One? Oh, I love it like hops 0:19:24.920 --> 0:19:29.919 and I'm si care just taste like grass? What are you? 0:19:32.359 --> 0:19:35.800 We're done? So the first thing you guys should tell 0:19:35.880 --> 0:19:40.000 us is which is your favorite? For me? Is number three? 0:19:40.119 --> 0:19:42.520 Just because I like the complexity of flavor. Three was 0:19:42.560 --> 0:19:46.919 a beverage too. I'm used to and I've probably had before. 0:19:46.960 --> 0:19:51.520 Maybe three. I really enjoyed one I did not like. 0:19:52.000 --> 0:19:55.000 I don't like those flavors. We have to reveal. We're 0:19:55.000 --> 0:19:58.600 almost at the reveal can to clarify. Your job as 0:19:58.640 --> 0:20:03.560 a company is to predict future products that people will 0:20:03.640 --> 0:20:07.200 enjoy and come back to. Yes, So in that regard, 0:20:07.520 --> 0:20:13.720 three was a winner. All right? Oh my god, tes us. 0:20:13.840 --> 0:20:15.639 So we had to do one product that was going 0:20:15.720 --> 0:20:18.159 to be optimal for both of you, and we got it. 0:20:18.160 --> 0:20:24.240 It was number three. They nailed it. You could say 0:20:24.240 --> 0:20:27.560 that we got flavor hacked up. Here. This this blue graph, 0:20:27.600 --> 0:20:29.840 this is saying that there's a you know, a seventy 0:20:29.920 --> 0:20:33.680 percent chance that you would give this a six? Did 0:20:33.680 --> 0:20:36.240 I give it a six? I think I did? You? Did? 0:20:36.280 --> 0:20:38.120 You both gave it a six? Um? So we were 0:20:38.119 --> 0:20:40.480 pretty confident on this, but we didn't have that level 0:20:40.520 --> 0:20:43.000 of confidence that we saw this. What I tasted is 0:20:43.040 --> 0:20:46.919 not something I ever predicted I would have liked, but 0:20:47.640 --> 0:20:50.639 it's absolutely something I will continue to think about. It 0:20:50.680 --> 0:20:53.240 was such a unique flavor and it's actually something I 0:20:53.240 --> 0:20:55.520 would buy. It's just that I had never tasted something 0:20:55.520 --> 0:20:57.800 like it before. So when I first tasted it, I 0:20:57.880 --> 0:21:01.760 was like, this is strange. In the case scenario, gastrographs 0:21:01.800 --> 0:21:05.560 AI can help companies create foods that satisfying more specific 0:21:05.640 --> 0:21:08.960 tastes and even bring people more joy, and that's good 0:21:08.960 --> 0:21:11.879 for business. Instead of making huge bets and trying to 0:21:11.920 --> 0:21:14.920 market a product to an entire country, a f S 0:21:14.960 --> 0:21:17.520 has created a way to make more specialized bets and 0:21:17.600 --> 0:21:20.800 help companies tap into those niches. And this isn't about 0:21:20.880 --> 0:21:24.720 AI reducing our experiences to data. AI is being used 0:21:24.720 --> 0:21:29.520 to change how we experience the world. More sleepwalkers after 0:21:29.560 --> 0:21:43.359 the break So, Karen, I would have been quite nervous 0:21:43.359 --> 0:21:45.800 to stand in the shoes of Jason and Ryan and 0:21:45.840 --> 0:21:48.359 the gastrograph team because I know that you're such a 0:21:48.440 --> 0:21:54.160 connoisseur of Seltzer. How do they do? They did really 0:21:54.200 --> 0:21:57.560 well actually, and I think it's important to mention that 0:21:57.680 --> 0:22:00.800 they weren't trying to create something they thought I would 0:22:00.880 --> 0:22:04.639 already like, like I love cranberry, right. They were trying 0:22:04.680 --> 0:22:08.680 to create something that I hadn't tasted before and also liked. 0:22:08.760 --> 0:22:11.439 So it was really difficult, and I think it also 0:22:11.560 --> 0:22:14.280 shows that there's a bit of reversal in the way 0:22:14.280 --> 0:22:17.560 that we do things. Companies have always used market research 0:22:17.680 --> 0:22:20.720 to predict consumer preference, but it's often based on things 0:22:20.720 --> 0:22:24.640 like focus groups or survey research. What we have now 0:22:25.080 --> 0:22:28.399 is massive amounts of data being funneled through an algorithm 0:22:28.640 --> 0:22:32.000 to deliver the perfect product for a very specific type 0:22:32.000 --> 0:22:37.679 of person exactly. That's the or yeah, age, demographic, socio 0:22:37.720 --> 0:22:42.640 economic race. They can target it to all these particular categories, 0:22:42.920 --> 0:22:46.639 and I think this is cool and also a little unsettling. 0:22:46.880 --> 0:22:48.840 I think as humans, we like to be in control, 0:22:48.920 --> 0:22:51.280 you know. I like to think that my preferences are 0:22:51.320 --> 0:22:55.040 just that, my own preferences, and this sort of up 0:22:55.200 --> 0:22:58.639