WEBVTT - L’Oréal and IBM: AI-Powered Beauty 0:00:02.080 --> 0:00:06.120 To understand why the cosmetics supergiant Lorel Group is teaming 0:00:06.200 --> 0:00:09.200 up with IBM, you must first take a closer look 0:00:09.280 --> 0:00:13.160 at its products. Take lipstick, for example, It's one of 0:00:13.160 --> 0:00:16.360 those things that seems straightforward, a waxy cylinder that you 0:00:16.440 --> 0:00:19.440 rub on your lips to turn them a different color. Easy, right, 0:00:20.000 --> 0:00:23.759 Well maybe not, as my colleague Lucy Sullivan found out 0:00:23.960 --> 0:00:27.040 when I sent her an assignment to Lorel's North America 0:00:27.120 --> 0:00:28.840 Research and Innovation Center. 0:00:29.920 --> 0:00:35.000 All right, I am reporting live from the Looreal visitor's 0:00:35.159 --> 0:00:39.040 parking lot. Malcolm told me that he would be sending 0:00:39.040 --> 0:00:44.919 me to Paris, France for this Looreal excursion, but instead 0:00:45.000 --> 0:00:48.360 I am in Clark, New Jersey. Pass a lot of 0:00:48.400 --> 0:00:51.760 strip malls on the way here. But to be fair 0:00:51.800 --> 0:00:55.600 to Clark, New Jersey and Lorel, this is a beautiful compound. 0:00:55.600 --> 0:00:58.800 It kind of looks like a spa. 0:00:59.000 --> 0:01:02.560 Lucy went into the center and was blown away. The 0:01:02.600 --> 0:01:07.840 facility houses about six hundred scientists and experts across skincare, makeup, fragrance, 0:01:07.920 --> 0:01:12.000 hair care, innovative packaging, and tech. It is one of 0:01:12.040 --> 0:01:15.319 the largest formulation lab spaces in the industry. It's the 0:01:15.360 --> 0:01:19.800 size of six basketball courts. The reason Loreel's facility is 0:01:19.840 --> 0:01:22.800 so big and has so many people is that everything 0:01:22.880 --> 0:01:25.720 Loriel does to bring a product to market happens here, 0:01:26.319 --> 0:01:30.920 from molecule discovery and product development to consumer testing. The 0:01:30.959 --> 0:01:35.040 center even has its own mini factory. My conception of lipstick, 0:01:35.360 --> 0:01:39.160 that it's just a waxy stick was plain wrong. Lipstick 0:01:39.400 --> 0:01:42.520 is a high performance product born from years of research, 0:01:42.880 --> 0:01:48.760 consumer insights, and precision science. Lipstick isn't simple. It's incredibly complex, 0:01:49.360 --> 0:01:51.840 and one of the main reasons it's so complex is 0:01:51.880 --> 0:01:54.720 just a nature of fashion trends. The kind of lipstick 0:01:54.760 --> 0:01:57.240 consumers want is constantly changing. 0:01:57.760 --> 0:01:59.880 A lot of our consumer insights with Floreal is like, 0:02:00.280 --> 0:02:02.160 where are consumers going in the future. 0:02:02.480 --> 0:02:06.440 This is Nadine Gomez. She's vice president for Loreel's research 0:02:06.480 --> 0:02:08.240 and innovation development team. 0:02:08.639 --> 0:02:11.040 Our chemists are working on five six years down the line. 0:02:11.240 --> 0:02:14.839 We predicted that consumers wanted more of a softer look 0:02:14.840 --> 0:02:15.760 on their lips as well. 0:02:15.880 --> 0:02:17.400 So how do you predict something like that. 0:02:17.960 --> 0:02:22.320 We see slow signals from fashion houses and social media 0:02:22.360 --> 0:02:24.239 and things like that. We kind of see that trend 0:02:24.320 --> 0:02:27.440 evolving a little bit, and then we know at five 0:02:27.480 --> 0:02:28.799 six years it's going to become. 0:02:28.560 --> 0:02:32.480 Big Lucy talked with her about the origins of one 0:02:32.520 --> 0:02:36.200 of their products, Mabelne matt Inc Liquid lipstick. 0:02:36.760 --> 0:02:39.080 Our competitors have two steps. The first step is a 0:02:39.120 --> 0:02:41.720 base go it's super opaque. You get the color and 0:02:41.760 --> 0:02:45.320 you get the maddy, but it's very, very drying ellips. 0:02:45.360 --> 0:02:47.360 You cannot wear that, honestly more than ten minutes. It 0:02:47.360 --> 0:02:50.120 feels like your lips are like aching at one point. 0:02:50.360 --> 0:02:51.960 So we had to develop a top cot and you'll 0:02:51.960 --> 0:02:53.720 see many of our competitors did the same thing. It's 0:02:53.720 --> 0:02:55.760 like a bomb. You put it on top, it's super comfortable, 0:02:56.400 --> 0:02:59.680 but we also noticed that consumers kind of get tired 0:03:00.120 --> 0:03:02.440 reapplying a bomb. So we're like, what can we do 0:03:02.560 --> 0:03:04.680 to create this two step into one step? 0:03:05.440 --> 0:03:08.280 So Loriel had a challenge, how do you make a 0:03:08.360 --> 0:03:12.960 comfortable liquid matt lipstick that doesn't require consumers to reapply 0:03:13.080 --> 0:03:16.560 a top layer of bomb. Solving this type of problem 0:03:16.760 --> 0:03:19.600 takes a lot of resources and a lot of expertise, 0:03:19.960 --> 0:03:24.160 and crucially, it takes time. Remember, Nadine said that working 0:03:24.200 --> 0:03:27.160 on a breakthrough product such as matt Inc can take 0:03:27.400 --> 0:03:31.240 years before it comes out. But can this process be 0:03:31.360 --> 0:03:36.240 accelerated taken further? Be even more sustainable. That's what IBM 0:03:36.280 --> 0:03:40.920 and Loreel are hoping to find out. My name is 0:03:40.920 --> 0:03:44.000 Malcolm Gladwell. You're listening to the latest episode of Smart 0:03:44.040 --> 0:03:47.000 Talks with IBM, where we offer our listeners a glimpse 0:03:47.040 --> 0:03:50.640 behind the curtain of the world of technology. In our 0:03:50.720 --> 0:03:54.240 last episode, we talked about how an AI assistant created 0:03:54.240 --> 0:03:58.360 with IBM Watson X helps future teachers practice responsive teaching 0:03:58.760 --> 0:04:04.200 by simulating class interactions with students. In this episode, we 0:04:04.360 --> 0:04:07.760 take you on an even more unexpected journey into the 0:04:07.760 --> 0:04:13.320 world of cosmetics, hair care, skincare, fragrance, makeup, and how 0:04:13.360 --> 0:04:17.359 a custom AI model could help Loriel's researchers shape the 0:04:17.400 --> 0:04:20.159 future of what we put on our faces every morning. 0:04:28.040 --> 0:04:30.159 I want to say on lipstick a moment longer to 0:04:30.240 --> 0:04:34.080 help illustrate what goes into Loriel's product development, and let's 0:04:34.080 --> 0:04:38.320 focus on matt inc lipstick. Loriel wanted to create something 0:04:38.520 --> 0:04:41.599 that was comfortable and could be applied in one step. 0:04:42.240 --> 0:04:44.279 So to go from two step to one step, we 0:04:44.320 --> 0:04:46.760 had to look cross functionally and try to figure out 0:04:46.760 --> 0:04:49.160 what can we bring into the product to make it 0:04:49.200 --> 0:04:52.400 more comfortable, and luckily we have many different types of 0:04:52.400 --> 0:04:53.320 products at Lorel. 0:04:53.720 --> 0:04:56.960 That's Alex Good, a senior chemist who leads the lip 0:04:56.960 --> 0:05:00.159 products team in North America. She says the trick to 0:05:00.240 --> 0:05:04.520 making matt incwork was finding an elastomer, a substance they 0:05:04.560 --> 0:05:06.560 were already using in foundation. 0:05:07.480 --> 0:05:10.360 We have this elastomer that can give you like more 0:05:10.360 --> 0:05:13.040 comfortable and make it feel like there's like something on 0:05:13.040 --> 0:05:14.000 your lips, like a cushion. 0:05:14.800 --> 0:05:18.320 She handed Lucy two jars. The first jar contained the 0:05:18.360 --> 0:05:20.839 former version of the product that was used in super 0:05:20.880 --> 0:05:24.280 State twenty four. By the way, this is exactly why 0:05:24.320 --> 0:05:26.800 I sent Lucy to the lab in my place the. 0:05:26.839 --> 0:05:30.000 Samples, and I actually have something for you to try here, 0:05:30.520 --> 0:05:34.599 so you can try this is what was the initial product. 0:05:34.920 --> 0:05:38.480 Okay, so this is like it sort of looks like okay, 0:05:38.480 --> 0:05:41.360 it is. It looks like vacline that has like more 0:05:41.400 --> 0:05:44.239 of a color. It's kind of a beige, looks like 0:05:44.279 --> 0:05:44.799 some skin. 0:05:44.960 --> 0:05:45.359 Okay. 0:05:45.600 --> 0:05:47.920 So this was from the two steps this would go 0:05:47.960 --> 0:05:55.080 on after Oh okay, right islet Okay, So. 0:05:55.080 --> 0:05:58.280 It feels like very wet. As you can see, it's 0:05:58.320 --> 0:06:01.279 kind of it's gonna absorb into your skin and leaves 0:06:01.800 --> 0:06:03.520 and then you're gonna feel the dryness. 0:06:03.120 --> 0:06:05.560 Of the product. Okay, so we're gonna move. 0:06:05.560 --> 0:06:07.640 From the clay product that you have on your hand 0:06:07.720 --> 0:06:09.640 now to the elastimmer or. 0:06:09.680 --> 0:06:10.520 You try half hour. 0:06:11.279 --> 0:06:14.760 This jar held the elastomer that Laurel had spent years 0:06:14.800 --> 0:06:15.800 developing in the lab. 0:06:16.800 --> 0:06:20.960 This one is it clear, looks like aqua for a 0:06:21.080 --> 0:06:23.040 much bligger and you. 0:06:23.040 --> 0:06:26.480 Can pay a physical layer that you're putting on your aid. 0:06:26.680 --> 0:06:30.920 Yeah, so that's much thicker. It kind of like clumps together. Yeah, 0:06:31.880 --> 0:06:35.640 it's more of a cloudy. It's less shimmery though that's intended. 0:06:35.880 --> 0:06:40.760 Yes, so this is a like a powder, this dispersed 0:06:40.880 --> 0:06:45.000 in dimetic code, and it creates like a comfort on 0:06:45.000 --> 0:06:47.080 your lips, a field like there's something there for a 0:06:47.120 --> 0:06:50.760 barrier to keep the film form on. And that's like 0:06:50.800 --> 0:06:53.680 the key ingredient that came from Foundation that we transferred 0:06:53.680 --> 0:06:56.719 into lipstick to give us this innovative product ahead of 0:06:56.760 --> 0:07:00.360 the market. Yeah, this is what gives it comfort. So 0:07:00.440 --> 0:07:03.200 the difference between super State twenty four and matt Inc 0:07:03.320 --> 0:07:05.800 is really the comfort. They both last a long time, 0:07:06.160 --> 0:07:08.800 but this matt Inc you don't have to apply the 0:07:08.839 --> 0:07:11.320 bomb over and over again, So you can apply matt 0:07:11.320 --> 0:07:12.440 ink once for the day and. 0:07:12.400 --> 0:07:16.200 You're good, right, Alex Good is under selling it here 0:07:16.840 --> 0:07:19.440 once for the day, and you're good. That's a liquid 0:07:19.480 --> 0:07:24.000 lipstick revolution. Literally millions of Loreal consumers around the world 0:07:24.240 --> 0:07:28.080 have worn matt ink. It's a blockbuster. It's also a 0:07:28.120 --> 0:07:33.200 marvel of science. The world's first liquid lipstick was developed 0:07:33.240 --> 0:07:35.520 in the nineteen thirties, and it was actually just a 0:07:35.600 --> 0:07:39.680 stain for your lips, barely counts as lipstick. Then came 0:07:39.760 --> 0:07:42.400 another wave of liquid lipstick when they were able to 0:07:42.440 --> 0:07:45.600 make it matt That was a two step version. It 0:07:45.600 --> 0:07:48.800 felt heavy on your lips. You had to keep reapplying 0:07:48.800 --> 0:07:53.600 the top coat. It was inconvenient. Lorel tackled that challenge 0:07:53.640 --> 0:07:56.679 in the lab, with chemists like Alex and Nadine leaving 0:07:56.720 --> 0:08:01.840 the charge their breakthrough matt Inc. But creating matt Inc 0:08:01.920 --> 0:08:05.040 took a long time, trial and error, the hard work 0:08:05.080 --> 0:08:10.000 of scientific experimentation. As Nadine told Lucy, the lipstick team 0:08:10.240 --> 0:08:13.160 had to put the new product to extensive tests. 0:08:13.600 --> 0:08:16.520 We do a very robustability system here. You know, we 0:08:16.560 --> 0:08:19.960 have color odor appearance. We monitor this in extreme conditions. 0:08:20.320 --> 0:08:22.920 We simulate a forty five degrees celsius and that can 0:08:22.960 --> 0:08:25.080 be something like a three year shelf life. I'm saying, 0:08:25.680 --> 0:08:28.600 we simulate your real life product. Like if you leave 0:08:28.640 --> 0:08:31.240 your lip gloss in the car in Arizona's one hundred 0:08:31.240 --> 0:08:33.040 and twelve degrees for three days, is it still going 0:08:33.080 --> 0:08:36.000 to perform? Is it gonna smell? Is it gonna look granted? 0:08:36.120 --> 0:08:37.199 Is it gonna change colors? 0:08:37.400 --> 0:08:38.160 We do all that. 0:08:38.800 --> 0:08:42.680 See what I mean. Lipstick is complex. Most people would 0:08:42.679 --> 0:08:45.600 never consider it a piece of technology, but one lip 0:08:45.600 --> 0:08:48.319 product has millions of data. 0:08:48.040 --> 0:08:50.439 Points, so much science behind. And you can see here 0:08:50.480 --> 0:08:52.280 how many scientists we have. You know, some of them 0:08:52.280 --> 0:08:56.120 have PhDs, some of them have masters degrees chemistry, biology, psychology. 0:08:56.160 --> 0:08:59.839 Also, when I first heard about this collaboration between LORI 0:09:00.120 --> 0:09:03.679 L and IBM, I was surprised. I thought, these are 0:09:03.720 --> 0:09:07.080 two very different companies. What do they really have in common? 0:09:07.880 --> 0:09:10.720 Meet you guys, Yeah. To find out, I went to 0:09:10.760 --> 0:09:13.840 the IBM Research Center outside New York City, which I 0:09:13.880 --> 0:09:15.600 have to say is one of the coolest buildings I've 0:09:15.640 --> 0:09:19.920 ever been in, A semi circular modernist masterpiece with a 0:09:19.920 --> 0:09:22.679 long curving wall of windows, looks like something out of 0:09:22.679 --> 0:09:25.560 a Stanley Kubrick movie. I was there to talk with 0:09:25.679 --> 0:09:29.880 two experts from research and innovation at Lorel, Methu Cassier 0:09:30.320 --> 0:09:35.079 and Gabriel Bertoli metthew is VP for Digital and Transformation. 0:09:35.600 --> 0:09:40.600 Gabriel is the Chief Digital Transformation Officer for Formulation. These 0:09:40.600 --> 0:09:43.680 are the people whose jobs are to oversee big changes 0:09:43.720 --> 0:09:46.840 within the company. And Methu told me to try on 0:09:46.880 --> 0:09:47.560 some lipstick. 0:09:48.000 --> 0:09:49.840 I'm gonna make you try this one. 0:09:50.240 --> 0:09:54.360 Okay, this is super stay viniting final inc. 0:09:54.559 --> 0:09:55.600 Yeah, so that's a glosse. 0:09:55.720 --> 0:09:57.040 Never in my life put on lipsey. You have no 0:09:57.080 --> 0:09:57.640 idea what I'm doing. 0:09:57.679 --> 0:09:59.440 You don't have to put it. You can try it virtu. 0:10:00.240 --> 0:10:03.640 Oh this may not be news to people who buy makeup, 0:10:03.880 --> 0:10:06.320 but it was news to me. You can try on 0:10:06.400 --> 0:10:11.600 Loreal products virtually. They call it augmented beauty. Oh my goodness. 0:10:12.280 --> 0:10:14.600 That is the strangest thing I've ever said. I look 0:10:14.679 --> 0:10:16.480 quite fetching, that's the way. 0:10:16.480 --> 0:10:16.920 It amazing. 0:10:18.160 --> 0:10:21.120 And I can just hit you can choose your color absolutely. 0:10:21.840 --> 0:10:24.199 So I'm on a little app it's looking at me 0:10:24.520 --> 0:10:26.839 and it's just showing me exactly how I would look 0:10:26.840 --> 0:10:29.360 with different shades of lipstick. So the odd idea of 0:10:29.400 --> 0:10:31.319 going into a store and trying on each one, you 0:10:31.400 --> 0:10:32.960 cannot do that from home, if you're not even at 0:10:33.000 --> 0:10:33.360 the store. 0:10:33.840 --> 0:10:36.840 Yeah, absolutely, that's all purpose. If you want to manage 0:10:36.880 --> 0:10:38.880 a trend, I would go for something more like pitch. 0:10:40.040 --> 0:10:41.240 You think I'm a peach person. 0:10:42.240 --> 0:10:42.520 I don't know. 0:10:43.000 --> 0:10:45.560 That looks I have to say that looks kind of natural. 0:10:45.600 --> 0:10:48.720 It just is enhanced. It's given me a boys share 0:10:48.760 --> 0:10:53.880 I would not otherwise have. This is why Loreel says 0:10:53.920 --> 0:10:58.480 it creates beauty products and beauty experiences. Loriel is a 0:10:58.480 --> 0:11:02.920 beauty tech company. Over the last decade, Laurel has seized 0:11:02.920 --> 0:11:07.319 the power of AI and more recently, generative AI technology 0:11:07.360 --> 0:11:11.600 has become a driving force alongside science and creativity. And 0:11:11.640 --> 0:11:15.520 while some of this digital technology is relatively new, Matthew 0:11:15.760 --> 0:11:19.480 helped me see that IBM and Lorel have always had 0:11:19.480 --> 0:11:20.240 a lot in common. 0:11:20.800 --> 0:11:23.920 I saw the original creator of Loyal, Jentulier, was a 0:11:23.960 --> 0:11:26.880 chemist in nineteen or nine, so one hundred and sixteen 0:11:26.920 --> 0:11:30.760 years ago, and he created this new air color type 0:11:31.000 --> 0:11:34.080 for the market in France, and then little by little, 0:11:34.400 --> 0:11:36.679 it has been always a very scientific company. So if 0:11:36.679 --> 0:11:39.120 you look a little bit at key facts, we invented 0:11:39.160 --> 0:11:43.040 sun filters in the nineteen thirties. There was a very 0:11:43.120 --> 0:11:45.760 very big milestone where we also invented not only product, 0:11:45.800 --> 0:11:49.400 but a reconstructed skin. So if you look at nineteen 0:11:49.800 --> 0:11:53.440 seventeen nine, we've been the created this reconstructed kin that 0:11:53.559 --> 0:11:55.839 helped us to go out of animal testing very fast 0:11:55.840 --> 0:11:57.920 and by the way, before the law even asked it 0:11:57.920 --> 0:12:01.880 to cosmetic companies. And then more recently, because it's a 0:12:01.960 --> 0:12:05.440 story of innovation, we launch on new molecules like one 0:12:05.440 --> 0:12:07.800 that you can find in laroche pose Milabi three, which 0:12:07.840 --> 0:12:11.280 is really helping people to find against some you know, 0:12:11.880 --> 0:12:13.920 spots that they could have on their skin. It's all 0:12:13.960 --> 0:12:16.200 about like big mountation, how to regulate it. 0:12:16.880 --> 0:12:20.440 Loreel and IBM were both started in the early twentieth century, 0:12:20.800 --> 0:12:24.400 Lorel in nineteen oh nine and IBM in nineteen eleven. 0:12:24.880 --> 0:12:28.760 Both companies have long standing histories of innovation, of using 0:12:28.800 --> 0:12:31.960 trial and error to improve everything they do. The two 0:12:31.960 --> 0:12:34.720 companies have been doing that in parallel for more than 0:12:34.800 --> 0:12:38.959 a century until recently. When does it start, When do 0:12:39.120 --> 0:12:40.720 Lorel and IBM start working together? 0:12:41.559 --> 0:12:44.440 So we started in twenty twenty three at the end 0:12:44.480 --> 0:12:46.600 of the year. But you know, really the discussion is 0:12:46.640 --> 0:12:51.240 really recent, absolutely, absolutely, it's really recent in reality, you know, 0:12:51.320 --> 0:12:54.640 I would say the first really interaction happened at the 0:12:54.640 --> 0:12:56.040 beginning of twenty twenty four. 0:12:56.840 --> 0:13:00.240 This is Gabriel Bertoli who I spoke to alongside too. 0:13:01.000 --> 0:13:04.199 What really played a key role here is we wanted 0:13:04.200 --> 0:13:08.400 to bring from a logic perspective to R and D together, 0:13:09.640 --> 0:13:13.080 which normally you know companies like us, you just go 0:13:13.200 --> 0:13:16.040 to a provider. You know it's a customer and the 0:13:16.080 --> 0:13:18.640 supplier and your work they delivered to you. Here the 0:13:18.679 --> 0:13:20.080 concept was totally different. 0:13:20.920 --> 0:13:24.360 Mid two said that the collaboration began with simple conversations. 0:13:24.720 --> 0:13:27.560 So if you look at the way IBM entered into 0:13:28.160 --> 0:13:32.400 loyal Labs, it's started by interviewing people what would help 0:13:32.440 --> 0:13:34.680 you to do your job? What is your business need? 0:13:35.200 --> 0:13:37.680 So it was, by the way, two months ago, a 0:13:37.880 --> 0:13:41.680 long series of interviews and from all the people around 0:13:41.720 --> 0:13:44.880 the world we have in research in Brazil, in India, 0:13:45.000 --> 0:13:49.720 in China, Japan, US, France of course, so we really 0:13:49.720 --> 0:13:51.320 want to make sure that at the end of the day, 0:13:51.679 --> 0:13:53.760 this new model, this new tool that we will give 0:13:53.800 --> 0:13:56.520 to people is really people centrick in the way that 0:13:56.600 --> 0:13:58.040 it selves their daily need. 0:13:58.679 --> 0:14:03.480 More. The point has leveraged technology for decades and accumulated 0:14:03.760 --> 0:14:09.080 amounted of scientific knowledge, everything from consumer aspirations and market trends, 0:14:09.400 --> 0:14:12.800 to the results of all the experiments conducted during product development, 0:14:13.400 --> 0:14:17.080 to which formulations melt in a hot car. It's hard 0:14:17.080 --> 0:14:20.880 to get your head around. Looreal isn't just a cosmetics company. 0:14:21.360 --> 0:14:23.640 It's a beauty data powerhouse. 0:14:24.560 --> 0:14:29.400 If we have sixteen thousand terabat of data coming from 0:14:29.760 --> 0:14:37.280 consumer insights, coming from market research coming from sales, well 0:14:37.440 --> 0:14:42.280 with the new technology, maybe by aligning those two and 0:14:42.400 --> 0:14:45.400 using best in class technology you can solve that problem. 0:14:45.480 --> 0:14:48.200 So you say you have sixteen terabytes of data, put 0:14:48.200 --> 0:14:50.160 that in perspective. How much data is that? 0:14:50.720 --> 0:14:51.040 Give me? 0:14:52.360 --> 0:14:56.320 This is one hundred year of Floreal data based on 0:14:56.520 --> 0:15:01.400 the last forty years of data the systems. So this 0:15:01.520 --> 0:15:03.400 is really I mean, we're talking about one hundred years 0:15:03.400 --> 0:15:06.440 of data that only Loreal have. Let's take the example 0:15:06.520 --> 0:15:08.880 of the ellipsis. I mean, you know, if ellipsex can 0:15:08.920 --> 0:15:13.080 be between twenty and thirty year ow material, each raw 0:15:13.120 --> 0:15:16.760 material will have I would say ten or fifteen way 0:15:16.960 --> 0:15:19.840 of doing things. 0:15:21.320 --> 0:15:23.800 Gabrielle is talking about how things used to be done. 0:15:24.120 --> 0:15:28.040 Researchers at Lorel needed roughly twenty five ingredients for a 0:15:28.040 --> 0:15:31.160 new lipstick formulation, but they have to choose from a 0:15:31.160 --> 0:15:35.480 pool of hundreds, if not thousands, of raw materials, And 0:15:35.560 --> 0:15:37.800 even after they settle on the ones they want, they 0:15:37.800 --> 0:15:40.200 have to figure out how much of each ingredient they 0:15:40.240 --> 0:15:44.600 need and in what form, what molecular weight, what combination. 0:15:45.360 --> 0:15:48.280 It's not just a math problem. It's a problem that 0:15:48.400 --> 0:15:55.920 requires balancing multiple perspectives safety, performance, quality, compliance standards, sustainability, 0:15:56.000 --> 0:15:59.720 and more. It can take years. But what if you 0:15:59.760 --> 0:16:03.560 could simulate hundreds of cars parked in a sweltering heat. 0:16:03.920 --> 0:16:06.480 What if you could do all those trials and errors 0:16:06.800 --> 0:16:10.840 virtually over and over and over again. What if instead 0:16:10.840 --> 0:16:14.640 of mixing materials together by hand, you could ask AI 0:16:14.760 --> 0:16:18.680 to predict what combinations might work best and then try 0:16:18.760 --> 0:16:19.720 those out first. 0:16:20.280 --> 0:16:24.440 This is ten on the power of twenty five. This 0:16:24.520 --> 0:16:29.000 is one hundred billion of years for a human to 0:16:29.120 --> 0:16:33.200 do a change in the formula or the possibility they have. 0:16:34.000 --> 0:16:39.040 You can only do this by using technology, power of 0:16:39.120 --> 0:16:40.640 technology and data that you have. 0:16:41.280 --> 0:16:44.560 This, Matthew says, is where IBM can come in to 0:16:44.640 --> 0:16:49.320 help take things further. Using artificial intelligence, IBM can help 0:16:49.360 --> 0:16:53.080 Loriel create a custom AI model that helps to crunch 0:16:53.120 --> 0:16:57.040 those numbers to be a companion to the researchers to 0:16:57.080 --> 0:16:58.160 give them superpowers. 0:16:58.400 --> 0:17:00.440 We don't want to replace the intuition of sent this. 0:17:00.520 --> 0:17:03.480 We just want to make sure that this intuition is 0:17:03.520 --> 0:17:07.840 really augmented by some calculation poor that, as Gabrielle said, 0:17:07.880 --> 0:17:10.400 then does all ten and the poor of twenty five 0:17:10.600 --> 0:17:13.840 solution and se probably try this one, this one, this one, 0:17:14.080 --> 0:17:16.760 it looks like a better solution, and then ultimately that's 0:17:16.800 --> 0:17:19.200 really the decision of the chemist to make it happen. 0:17:21.880 --> 0:17:24.960 Well. To make a predictive AI model that can give 0:17:25.080 --> 0:17:29.040 Loreal researchers those superpowers, you'd need that mountain of data, 0:17:29.600 --> 0:17:33.800 years worth of laboratory testing and all Loreal's data digitized 0:17:33.960 --> 0:17:37.960 and AI ready. You need to train artificial intelligence on 0:17:38.160 --> 0:17:41.000 everything the company has already done in order for it 0:17:41.119 --> 0:17:42.639 to predict what it could do. 0:17:43.160 --> 0:17:47.440 Loriial has one hundred years a course of data, fifty 0:17:47.560 --> 0:17:49.400 years of digitized EGGDA. 0:17:49.960 --> 0:17:53.600 This is Miriam Ashuri, Senior director of Product Management for 0:17:53.720 --> 0:17:57.159 IBM Watson X. Lorel has the data, and part of 0:17:57.200 --> 0:18:00.120 IBM's job is to help put that data to work, 0:18:00.520 --> 0:18:04.800 which involves ensuring data quality. Mariam talked about the concept 0:18:05.040 --> 0:18:06.480 of AI ready data. 0:18:07.160 --> 0:18:10.000 The sole purpose of this data engineering pipeline is to 0:18:10.119 --> 0:18:14.600 clean the data and we call them AI ready data 0:18:14.800 --> 0:18:18.600 makes them ready to be consumed by AI. So basically 0:18:18.640 --> 0:18:22.399 looking into biases and the data to fix the distribution, 0:18:22.600 --> 0:18:25.960 looking into guard brains that we are putting into place 0:18:26.000 --> 0:18:28.560 in terms of removing personal information. 0:18:29.440 --> 0:18:32.480 Marriam that explained that a custom model like the one 0:18:32.560 --> 0:18:35.760 IBM is creating with Loril can be more efficient and 0:18:35.920 --> 0:18:39.200 targeted than the larger general purpose AI models. 0:18:39.400 --> 0:18:42.760 You've heard about large language models. The reason that they 0:18:42.800 --> 0:18:46.359 call them large language model is they are exposed into 0:18:47.760 --> 0:18:51.800 really large amount of data. So the larger the model, 0:18:51.840 --> 0:18:54.840 the more take of all the models are, but also 0:18:55.280 --> 0:18:59.440 the larger computed requires that translates and increase carbon footprint 0:18:59.520 --> 0:19:04.440 and entergy consumption, that translatestan increase latency that's your response 0:19:04.480 --> 0:19:08.520 time that translatestand increased costs. So you started seeing that 0:19:09.240 --> 0:19:14.320 enterprises started grabbing a much smaller model customize it on 0:19:14.400 --> 0:19:18.680 their proprietary data that's the data, their DOMAINO specific data, 0:19:18.800 --> 0:19:22.600 or the data about their users to create something differentiated 0:19:23.200 --> 0:19:26.359 that is applicable to a real world use case but 0:19:26.560 --> 0:19:30.159 also delivers the performance that they needed for a fraction 0:19:30.280 --> 0:19:33.040 of the costs, and that's why there's been a lot 0:19:33.080 --> 0:19:37.439 of push around using custom models versus very large general 0:19:37.480 --> 0:19:38.560 purpose models. 0:19:39.160 --> 0:19:42.560 So how is a custom model created? Miriam says, you 0:19:42.640 --> 0:19:45.720 start with the base model. Imagine you're buying a car. 0:19:46.200 --> 0:19:48.359 You could get a minivan, or a sedan or a 0:19:48.359 --> 0:19:51.240 sports car, and then you get to customize it. You 0:19:51.240 --> 0:19:54.520 could add a sunroof, leather seats, or a rearview camera. 0:19:55.040 --> 0:19:57.080 Turns out you could do the same thing with your 0:19:57.119 --> 0:19:59.960 AI model. You pick a base and then you cut 0:20:00.000 --> 0:20:03.159 customize it. You tune it on the data unique to 0:20:03.200 --> 0:20:04.080 your organization. 0:20:04.520 --> 0:20:08.360 We do believe that one model doesn't fit all use cases. 0:20:09.040 --> 0:20:13.000 You want to truly have access to any model anywhere, 0:20:13.080 --> 0:20:17.480 and by any model anywhere, I really mean any model anywhere, 0:20:17.560 --> 0:20:22.439 open source, proprietary, low call out your machine wherever the 0:20:22.480 --> 0:20:25.919 model is. You want to host it yourself, because then 0:20:26.119 --> 0:20:28.960 you would be able to take advantage of the best 0:20:29.000 --> 0:20:31.840 of the technology at any point and pick the right 0:20:31.920 --> 0:20:33.399 model for the target use case. 0:20:33.760 --> 0:20:37.159 So a custom model tuned on Lorel's data would be 0:20:37.240 --> 0:20:41.399 more targeted and efficient than a general purpose model. It 0:20:41.400 --> 0:20:46.600 would understand the researchers world and provide transparency into its workings. 0:20:46.960 --> 0:20:49.800 That's part of the magic. And what could a custom 0:20:49.920 --> 0:20:53.880 AI foundation model do for a company like lorel. 0:20:54.400 --> 0:20:59.840 Accord with more contain the complexity of the formulation. 0:21:00.520 --> 0:21:04.600 That's Gillomme lais Moline, an IBM distinguished engineer and one 0:21:04.640 --> 0:21:06.800 of the people working on the AI model. 0:21:07.040 --> 0:21:12.040 And to hype ower the formulator to go not only 0:21:12.359 --> 0:21:17.200 past but also I would say, be able to include 0:21:17.320 --> 0:21:21.520 more complexity or so in the formulation, more personalization, more 0:21:21.680 --> 0:21:26.639 certain ability, better selected ingredient. So it's really a tool 0:21:26.720 --> 0:21:30.560 to help them and to also help them to unniche 0:21:30.720 --> 0:21:31.560 the creativity. 0:21:34.640 --> 0:21:38.360 Kayomi is saying that with its custom AI model, Loreel 0:21:38.400 --> 0:21:42.000 can improve every step of its product development pipeline, make 0:21:42.080 --> 0:21:45.760 the process faster and more sustainable. But he's also saying 0:21:46.000 --> 0:21:48.880 that the model could help Loriel create something that's never 0:21:48.920 --> 0:21:56.120 been done before. What could that product be? So I'm 0:21:56.119 --> 0:21:57.800 mourning you with that. All my questions are going to 0:21:57.800 --> 0:21:58.760 be really dumb. 0:21:59.560 --> 0:22:01.240 Okay, no, please by all me. 0:22:03.000 --> 0:22:05.159 To find out what people at Loril are dreaming of. 0:22:05.640 --> 0:22:09.520 I spoke with Trisha Iyagari, global general manager at Loriel's 0:22:09.520 --> 0:22:13.120 Mabeline brand and they asked her about her own dreams 0:22:13.240 --> 0:22:16.280 and how technology and science could help bring those dreams 0:22:16.600 --> 0:22:19.520 into the world. Do you have a secret wish list 0:22:19.800 --> 0:22:23.000 of things you think that this partnership could produce, Like, 0:22:23.119 --> 0:22:25.200 is there a product out there that's been technically too 0:22:25.200 --> 0:22:28.920 difficult that you think could be a worthy target? 0:22:29.240 --> 0:22:31.280 There is one that I think could be really amazing. 0:22:31.480 --> 0:22:31.800 What's that? 0:22:32.600 --> 0:22:35.479 So? Shine products in general are harder to create, and 0:22:35.560 --> 0:22:42.159 we're unable to create a shiny, long wearing eyeshadow. So 0:22:42.320 --> 0:22:44.920