WEBVTT 00:00:03.520 --> 00:00:04.240 I'm Farabostik. 00:00:04.320 --> 00:00:06.960 I'm just here to facilitate this conversation. 00:00:07.040 --> 00:00:09.520 We have a wonderful panel of expert 00:00:09.600 --> 00:00:14.680 panelists who are going to talk us through these topics related to emotive AI. 00:00:14.760 --> 00:00:18.480 But the way that I wanted to just ground the conversation is 00:00:18.560 --> 00:00:23.080 a little bit about getting underneath the hype, because I think you may have all 00:00:23.160 --> 00:00:28.800 seen these headlines about by next year, we'll see something like $200 billion 00:00:28.880 --> 00:00:31.960 in investment, according Goldman Sachs, in AI. 00:00:32.040 --> 00:00:35.600 We're seeing more and more of these generative AI tools like ChatGPT 00:00:35.680 --> 00:00:40.440 and Claude and Midjourney, which is one of my favorite things to play with. 00:00:40.520 --> 00:00:43.720 Actually, I just like to watch other people make things in Midjourney. 00:00:43.800 --> 00:00:45.600 Get yourself a Midjourney account just 00:00:45.600 --> 00:00:47.320 to watch what everyone else is making in the Discord. 00:00:47.400 --> 00:00:50.240 It is a good time. It's a really good time. 00:00:50.320 --> 00:00:55.120 But you also have a sense that things are really being disrupted. 00:00:55.200 --> 00:00:59.960 Hi, Vinnie. One of the ways that it's being disrupted, 00:01:00.040 --> 00:01:04.720 I think, really was brought home to me in a headline recently about Tyler Perry, 00:01:04.800 --> 00:01:09.680 the filmmaker, deciding not to build a movie studio because he'd seen a demo 00:01:09.760 --> 00:01:13.640 of Sorra from OpenAI and was just like, That's it. 00:01:13.720 --> 00:01:15.200 That's what we're going to be doing now. 00:01:15.280 --> 00:01:16.920 I think one of the questions that we're 00:01:17.000 --> 00:01:21.000 going to talk about tonight is the role of generative AI in creativity more 00:01:21.080 --> 00:01:26.080 generally, and also to really dig underneath what emotive AI means and why 00:01:26.080 --> 00:01:28.440 it's a topic that we should all be concerned about right now. 00:01:28.520 --> 00:01:30.120 I think it starts to bring things 00:01:30.200 --> 00:01:36.240 into a more concrete and relatable space as opposed to some of the big, 00:01:36.240 --> 00:01:37.920 broad topics that so frequently get talked about. 00:01:38.000 --> 00:01:42.600 What I want to do first is introduce our panelists. 00:01:42.680 --> 00:01:45.720 We have a wonderful group of people here, 00:01:45.800 --> 00:01:51.640 and each one of them comes from a slightly different expertise and point of view. 00:01:51.720 --> 00:01:55.600 To begin with, I want to introduce... 00:01:55.680 --> 00:01:57.520 I'm going to start actually in the middle and work my way out. 00:01:57.600 --> 00:02:00.040 I'm going to start with Carly, Carly Burton. 00:02:00.120 --> 00:02:03.800 Who is Director of Gen AI Product Design at Metta. 00:02:03.800 --> 00:02:05.800 We're going to have a lot to talk about there. 00:02:05.880 --> 00:02:08.640 We have Syra Jaisani on my left, 00:02:08.720 --> 00:02:12.560 who is Deputy Executive Director at the Data and Trust Alliance, 00:02:12.640 --> 00:02:16.280 which is incredibly important for us to get our heads around in this conversation. 00:02:16.360 --> 00:02:19.160 Next to her is Alejandro Matamal-Ortiz, 00:02:19.240 --> 00:02:22.800 who is co-founder and Chief Digital Officer at Runway, 00:02:22.880 --> 00:02:27.240 which is doing really cool stuff with ML and has been at this for a while. 00:02:27.320 --> 00:02:29.520 We'll be talking about storytelling with him. 00:02:29.600 --> 00:02:34.840 At At the very end, Joe Preston, who is VP of Product and Design at Intuit. 00:02:34.920 --> 00:02:36.600 Last, but certainly not least, 00:02:36.680 --> 00:02:40.560 is Maz Karimian, who is Head of Strategy at our Hosts Us, Two. 00:02:40.640 --> 00:02:42.360 We are going to actually... 00:02:42.440 --> 00:02:45.680 Why don't we just welcome the panelists one more time? 00:02:45.760 --> 00:02:50.000 All right. So seal broken. 00:02:50.080 --> 00:02:51.400 We can get on with it. 00:02:51.480 --> 00:02:52.760 Maz, this is all your fault. 00:02:52.840 --> 00:02:56.520 So why don't you talk about why you wanted to talk about Emotev AI? 00:02:56.600 --> 00:02:57.920 Why does it matter now? 00:02:58.000 --> 00:03:01.040 And what do you mean? Yeah, totally. 00:03:01.040 --> 00:03:04.640 Well, I'm going to begin by talking about my favorite topic, which is me. 00:03:04.720 --> 00:03:05.880 So I'm Maz. 00:03:05.960 --> 00:03:08.840 I head up US strategy for us, too. 00:03:08.920 --> 00:03:12.840 What that means is a hybrid of product and business strategy. 00:03:12.920 --> 00:03:17.000 And a lot of what I do is focused on humanizing emerging technologies, 00:03:17.080 --> 00:03:21.200 taking them from capabilities to useful applications. 00:03:21.280 --> 00:03:23.440 And a big lens that we apply to that, 00:03:23.480 --> 00:03:26.880 an emotive lens that we apply to that, is what we call play thinking, 00:03:26.960 --> 00:03:32.720 which is essentially a design ethos informed by our game design DNA. 00:03:32.720 --> 00:03:33.840 We're a product design studio, 00:03:33.900 --> 00:03:36.400 but we have a lot of game design experience in DNA, 00:03:36.480 --> 00:03:40.520 and we seek to bring those very emotion-centric, 00:03:40.600 --> 00:03:45.040 game design-informed principles and practices to product design. 00:03:45.120 --> 00:03:47.320 Where that links into emotive AI, 00:03:47.400 --> 00:03:50.400 of course, is the way in which we have this 00:03:50.480 --> 00:03:56.160 focus on AI, this discourse around AI as being very functionally maximizing. 00:03:56.240 --> 00:03:59.280 It automates processes. 00:03:59.280 --> 00:04:01.400 It might even automate or accelerate people. 00:04:01.480 --> 00:04:06.480 But we think that there's a whole side to AI, an emotive side to AI that we're 00:04:06.480 --> 00:04:08.360 really just beginning to scratch the surface of. 00:04:08.440 --> 00:04:09.480 What do I mean by that? 00:04:09.560 --> 00:04:11.960 We have folks like Alejandro who's here 00:04:12.040 --> 00:04:16.240 and tell you more about what Runway ML is doing in the text to video and video 00:04:16.320 --> 00:04:20.280 to video and A to B space, all these kinds of things. 00:04:20.360 --> 00:04:24.800 But what I'm getting at is that there are really, really emotive outputs from AI. 00:04:24.880 --> 00:04:27.120 These things can make us feel things 00:04:27.120 --> 00:04:29.480 and they can help us make other people feel things. 00:04:29.560 --> 00:04:31.000 Why does that matter? 00:04:31.080 --> 00:04:36.080 Fundamentally, because we as people make choices about the products and services we 00:04:36.160 --> 00:04:40.800 buy that we bring into our lives, not just on the basis of their function, 00:04:40.880 --> 00:04:44.880 but in terms of what they make us feel and how they satisfy us emotionally. 00:04:44.960 --> 00:04:49.440 So that's why we want to explore emotive AI as a topic today. 00:04:49.520 --> 00:04:54.120 I think the place I want to go next is maybe to you, Carly, 00:04:54.200 --> 00:04:59.120 to talk about the context for these interactions, because I think 00:04:59.200 --> 00:05:02.080 we We have certainly gone through this before. 00:05:02.080 --> 00:05:05.120 I'm old enough to remember when the hot new thing was Facebook, 00:05:05.200 --> 00:05:09.480 and we were talking about participatory brands and consumer-generated content. 00:05:09.560 --> 00:05:12.320 I think context has always mattered 00:05:12.320 --> 00:05:15.480 and sometimes gets overlooked in these rushes to do new problems. 00:05:15.560 --> 00:05:18.880 And so maybe you could talk a bit about how brands should think about 00:05:18.960 --> 00:05:23.960 and businesses should think about the context of the interaction as they 00:05:23.960 --> 00:05:27.000 start to think about more emotive interactions with generative AI. 00:05:27.080 --> 00:05:28.600 Hi, everybody. 00:05:28.680 --> 00:05:30.360 I think I'll start with a story to answer 00:05:30.440 --> 00:05:36.000 that question is about a decade ago, I did a global ethnography with digital 00:05:36.080 --> 00:05:40.200 natives to understand what is it that they expect out of technology. 00:05:40.280 --> 00:05:42.560 And some of the stories that they shared 00:05:42.640 --> 00:05:46.880 were things that I just never thought someone would have that close 00:05:46.960 --> 00:05:50.480 of a relationship, having grown up not with a computer 00:05:50.560 --> 00:05:54.440 myself, but their relationship was so different. 00:05:54.520 --> 00:05:57.360 And so I ended up writing about artificial 00:05:57.440 --> 00:06:01.640 empathy, and I thought, well, gosh, you need to have have so much data 00:06:01.720 --> 00:06:05.440 in order to enable that sense of knowing and understanding. 00:06:05.520 --> 00:06:09.960 And so when I think about the context in which Emotive AI needs to present 00:06:10.040 --> 00:06:13.120 itself, I think first about, what's its relationship? 00:06:13.200 --> 00:06:15.400 So it's an intelligent actor. 00:06:15.480 --> 00:06:19.720 It's going to change how you're going to engage with a product or a service. 00:06:19.800 --> 00:06:23.120 And so there's a framework that you could consider. 00:06:23.200 --> 00:06:24.560 What's its role? 00:06:24.640 --> 00:06:26.400 What's my reaction to it? 00:06:26.480 --> 00:06:28.480 What's my relationship to it? 00:06:28.560 --> 00:06:30.320 And how does it change my routine? 00:06:30.400 --> 00:06:32.280 So these four different R's. 00:06:32.360 --> 00:06:37.680 And the reason I think about that is you need it to deliver on how you expect it 00:06:37.680 --> 00:06:40.240 to show up within the context that you're designing for. 00:06:40.320 --> 00:06:45.280 So I'm sure there's a lot of product designers or product managers or even 00:06:45.280 --> 00:06:48.360 technologists in the room, is that you're always thinking about, 00:06:48.360 --> 00:06:50.880 okay, what's the value proposition that I'm going to deliver? 00:06:50.960 --> 00:06:52.600 And that means that there's a specific set 00:06:52.680 --> 00:06:54.880 of needs, someone's coming to your product, 00:06:54.960 --> 00:06:59.320 and then you're going to deliver these capabilities, and the AI is one of them. 00:06:59.400 --> 00:07:03.560 Well, How do I get it to emote the type 00:07:03.640 --> 00:07:06.760 of experience and have the right type of engagement? 00:07:06.840 --> 00:07:10.600 And then when I think about context, I think about public versus private. 00:07:10.680 --> 00:07:12.560 So if I'm going to have a conversation 00:07:12.640 --> 00:07:16.480 and something deeply knows me, maybe I just want that to be between me 00:07:16.560 --> 00:07:19.920 and that AI, and maybe I want it to behave in a certain way. 00:07:20.000 --> 00:07:23.440 Well, in order for it to do that, it needs to know quite a lot about me. 00:07:23.520 --> 00:07:26.960 Then if I think about public instances, 00:07:27.040 --> 00:07:30.920 what's its role, relationship, and the reactions that are going to get 00:07:31.000 --> 00:07:34.280 created of the individuals maybe in that group? 00:07:34.360 --> 00:07:38.040 One of the jokes that I was actually making with my neighbors recently was 00:07:38.120 --> 00:07:42.000 like, I don't want it to replace 00:07:42.080 --> 00:07:46.360 the relationship that I have with you as I've gotten to know everyone 00:07:46.440 --> 00:07:49.720 in the neighborhood, but I would like for it to help me get 00:07:49.800 --> 00:07:53.960 to know new things about you so that I can figure out what this crazy pickleball is. 00:07:54.040 --> 00:07:56.240 That sounds great. 00:07:56.320 --> 00:07:58.720 So, yeah, it's a long-winded way of saying 00:07:58.800 --> 00:08:04.440 that I think the younger generation thinks about artificial empathy, 00:08:04.520 --> 00:08:11.600 and so the context in which AI can show up is quite broad, but how it should show up, 00:08:11.680 --> 00:08:15.920 what it can do is very dependent on the underlying technology, 00:08:16.000 --> 00:08:21.560 that being the data and then the ways that you shape it, and then the framework 00:08:21.560 --> 00:08:23.080 you put around it when you bring it to life. 00:08:23.160 --> 00:08:26.920 I think the other interesting challenge then, if we have brand managers 00:08:27.000 --> 00:08:31.320 in the room or people who work in that space, is wanting to present 00:08:31.400 --> 00:08:34.640 a brand personality through these interactions. 00:08:34.720 --> 00:08:36.880 I think when we were talking a couple 00:08:36.960 --> 00:08:41.200 of weeks ago now, we were talking about the difference between you have 00:08:41.200 --> 00:08:42.840 a personality, your personality is your personality. 00:08:42.920 --> 00:08:45.000 But depending on the situation, 00:08:45.080 --> 00:08:49.120 your tone, your word choice, whatever might be different. 00:08:49.120 --> 00:08:52.560 I'm curious if you have any thoughts about how brands might think of that new 00:08:52.640 --> 00:08:56.800 potential for dynamism where the agent is acting on its behalf and trying to live 00:08:56.880 --> 00:09:01.920 the brand's values and all of that, but still being in tune 00:09:02.000 --> 00:09:05.960 with the moment and what's needed there from a voice point of view. 00:09:06.040 --> 00:09:06.400 Yeah. 00:09:06.480 --> 00:09:10.040 I mean, contextual reasoning is a high compute use case. 00:09:10.120 --> 00:09:13.040 I think that when you're talking about 00:09:13.120 --> 00:09:18.160 dynamicism, it's how dynamic can these agentic behaviors be expressed. 00:09:18.240 --> 00:09:22.840 What you're mentioning is we all talk about personality, right? 00:09:22.920 --> 00:09:24.760 And personality is your voice. 00:09:24.840 --> 00:09:28.480 It's innately who you are, but your tone fluctuates. 00:09:28.560 --> 00:09:30.360 Today, maybe I'm a little bit I'm 00:09:30.440 --> 00:09:35.520 lethargic, and so I'm going to talk slower, or that changes. 00:09:35.600 --> 00:09:38.280 But sometimes whenever these models are 00:09:38.360 --> 00:09:43.400 trained to express as a brand, it wants to reflect that, right? 00:09:43.480 --> 00:09:50.520 And so how How somebody trains that, it needs to be able to express. 00:09:50.600 --> 00:09:52.240 But then can it be dynamic? 00:09:52.320 --> 00:09:56.240 And that contextualizing, can it know through multi-term 00:09:56.320 --> 00:10:00.040 conversations to pick up and store that memory and make those changes? 00:10:00.120 --> 00:10:02.120 These are high compute use cases. 00:10:02.200 --> 00:10:08.760 And so I think that there's the ambition and intent of brand expression and voice 00:10:08.840 --> 00:10:13.560 and making sure that you're able to create these agents that reflect that. 00:10:13.640 --> 00:10:18.520 But then there's also, we as humans are incredibly dynamic. 00:10:18.600 --> 00:10:19.680 We are unpredictable. 00:10:19.760 --> 00:10:21.840 We change. 00:10:21.920 --> 00:10:27.320 And can those models change for that context and express in the right way? 00:10:27.400 --> 00:10:28.080 Yeah. 00:10:28.160 --> 00:10:30.680 I wonder then, and maybe Now, as I'll start with you, 00:10:30.760 --> 00:10:36.080 but anyone feel free to jump in on this, but I'm wondering about how more emotive 00:10:36.160 --> 00:10:39.920 AI experiences could really fundamentally change 00:10:39.920 --> 00:10:41.960 the relationship between the brand and the customer. 00:10:42.040 --> 00:10:48.880 You can imagine suboptimal scenarios. 00:10:48.960 --> 00:10:54.160 Here I'm even thinking this happens even with humans in the situation where you're 00:10:54.160 --> 00:10:57.640 calling into a customer care center because you're freaked out about a bill 00:10:57.640 --> 00:11:00.480 you got that you weren't expecting, and the person on the other line is not 00:11:00.560 --> 00:11:03.400 meeting you where you are emotionally in the call. 00:11:03.480 --> 00:11:06.840 They're super chipper and you're having the worst day of your life. 00:11:06.920 --> 00:11:08.920 Those things could really matter. 00:11:08.920 --> 00:11:12.440 It seems like there's more opportunity to flex up and down those kinds of things. 00:11:12.440 --> 00:11:15.280 But I'm wondering how you're thinking about that from a design point of view. 00:11:15.360 --> 00:11:20.080 Yeah, I think to put it really crudely, I think right now we have relationship 00:11:20.080 --> 00:11:23.040 with apps, and we're going to have relationships with agents. 00:11:23.120 --> 00:11:24.680 We're not going to call them that, I think. 00:11:24.760 --> 00:11:26.920 But where the difference will come 00:11:26.920 --> 00:11:30.400 in and where I think we'll really feel it, is that when you think about how you 00:11:30.480 --> 00:11:33.800 engage with a brand today, it is forgetful. 00:11:33.880 --> 00:11:35.920 It is not continuous. 00:11:36.000 --> 00:11:38.720 At best, when you engage with a brand 00:11:38.800 --> 00:11:42.320 through an app or even in a physical location or what have you, 00:11:42.400 --> 00:11:45.280 the best that you can hope for is that they're going to pull up your file 00:11:45.360 --> 00:11:49.080 and they're going to maybe hopefully remember what it is that you 00:11:49.160 --> 00:11:52.520 bought that you're having an issue with or that you bought last time, 00:11:52.520 --> 00:11:56.200 and maybe you want to buy something or change something about it, what have you. 00:11:56.280 --> 00:12:00.800 When we think about the possibilities with AI, and especially emotive AI, 00:12:00.880 --> 00:12:05.720 the ability of AI to understand not just your stated preferences, your purchases, 00:12:05.800 --> 00:12:11.800 and what have you, but your personal maybe style in the context of clothing or 00:12:11.880 --> 00:12:14.640 cosmetics, your financial goals and your risk 00:12:14.720 --> 00:12:18.600 aversion or your risk appetite in the context of finance or any number 00:12:18.680 --> 00:12:24.040 of psychoemotional factors in the case of health and wellness and fitness, 00:12:24.120 --> 00:12:29.120 we have that ability to not just remember, but adapt to you and your experience as 00:12:29.200 --> 00:12:33.800 a customer on on the basis of what we know about you at every single moment 00:12:33.880 --> 00:12:37.880 in your experience as a customer, not just at those moments when you get put 00:12:37.880 --> 00:12:40.120 in contact with a human who can review your case. 00:12:40.200 --> 00:12:43.040 And maybe, hopefully, sift through the information and maybe, 00:12:43.040 --> 00:12:46.000 hopefully, make a connection that is relevant and meaningful to you. 00:12:46.080 --> 00:12:48.000 What we're getting at here is that agents 00:12:48.080 --> 00:12:53.120 can feel that connective tissue that always on relationship to the point 00:12:53.140 --> 00:12:56.640 Carly was making about having a real relationship dynamic with a brand. 00:12:56.720 --> 00:12:58.280 That is completely different than what we 00:12:58.360 --> 00:13:01.240 face now, which is where we have relationships with apps. 00:13:01.320 --> 00:13:03.920 We will soon have relationships with agents. 00:13:04.000 --> 00:13:06.320 That does seem like... Oh, go ahead, Carly. 00:13:06.400 --> 00:13:09.840 I was just going to say, Enter the data challenge, right? 00:13:09.920 --> 00:13:14.000 And user's perception of what you explain, right? 00:13:14.080 --> 00:13:18.080 It's like, Do people want their data used in that way as well? 00:13:18.160 --> 00:13:21.280 Yeah, I think that's always... 00:13:22.120 --> 00:13:25.360 Years ago when I was at Holland Partners and we were talking about participatory 00:13:25.440 --> 00:13:30.000 brands, a lot of the conversation was like, How intimate of a relationship do 00:13:30.080 --> 00:13:33.240 you really want to have with every brand you interact with. 00:13:33.320 --> 00:13:34.960 Some of them, you just want to give them 00:13:34.960 --> 00:13:36.560 your money, they give you the product, it's over. 00:13:36.640 --> 00:13:38.720 That's the end of the interaction. 00:13:38.800 --> 00:13:42.080 But others do instill more 00:13:42.160 --> 00:13:47.080 of an emotional interaction or have a more sensitive interaction. 00:13:47.080 --> 00:13:48.680 Maybe this is a good place for you to come 00:13:48.760 --> 00:13:51.280 in, Joan, talking about a couple of things. 00:13:51.360 --> 00:13:54.120 One is your work is dealing with what 00:13:54.120 --> 00:13:57.560 could be extremely sensitive financial information about your customers. 00:13:57.640 --> 00:13:59.240 The other thing that's interesting is it's 00:13:59.320 --> 00:14:03.440 a portfolio of products, and not all of them are even in financial services. 00:14:03.440 --> 00:14:04.920 But maybe talk a bit about how you think 00:14:05.000 --> 00:14:09.040 about brand and context and interaction across a portfolio of brands. 00:14:09.120 --> 00:14:14.160 Yeah, I love this conversation because I think once we've crossed that line of it's 00:14:14.160 --> 00:14:16.560 not perceptible to humans anymore, which I think we have. 00:14:16.640 --> 00:14:19.960 I think the designers role in most 00:14:20.040 --> 00:14:24.000 companies is asking, should we be doing this or not? 00:14:24.080 --> 00:14:27.480 Because I think we can't really see what 00:14:27.560 --> 00:14:31.680 the implications are to some of this stuff, especially to some of our audiences 00:14:31.760 --> 00:14:37.800 or customers that will be extremely vulnerable or able to be manipulated. 00:14:37.880 --> 00:14:40.240 From a financial context, 00:14:40.320 --> 00:14:46.640 why emotional AI is really interesting is because if you think about your financial 00:14:46.720 --> 00:14:51.320 life and how you start it out and how you want to be explained things in very simple 00:14:51.400 --> 00:14:58.160 terms and maybe in a tone that manages your confidence and your trust. 00:14:58.240 --> 00:15:00.880 And then as your financial life 00:15:00.880 --> 00:15:04.520 transcends, maybe you become a business owner, maybe you become very savvy. 00:15:04.600 --> 00:15:07.280 It's a really opposite end of the spectrum 00:15:07.360 --> 00:15:13.280 with how you want to be communicated to your financial savviness, all that thing. 00:15:13.360 --> 00:15:17.040 But again, I go back to we're crossing 00:15:17.120 --> 00:15:20.640 this interesting chasm, I think, right now. 00:15:20.720 --> 00:15:23.280 Just today, I was on my Instagram. 00:15:23.360 --> 00:15:28.480 My teenage girl is falling in love with the ChatGPT voice. 00:15:28.760 --> 00:15:30.240 Good voice. 00:15:30.320 --> 00:15:33.240 Yeah, I was like, wow, I never would have considered that. 00:15:33.320 --> 00:15:35.480 And I'm sure OpenAI didn't consider that. 00:15:35.560 --> 00:15:39.160 And I don't know how that will play out in the years to come. 00:15:39.240 --> 00:15:43.160 But it's incredible opportunities. 00:15:43.240 --> 00:15:44.320 But at the same time, 00:15:44.400 --> 00:15:49.160 I think this line we're crossing right now is really telling for humanity. 00:15:49.240 --> 00:15:51.560 Yeah. I think the other interesting opportunity 00:15:51.640 --> 00:15:54.160 here, and maybe, Alejandro, you can comment on this, 00:15:54.240 --> 00:15:59.440 is thinking about what this means beyond just the tone and manner 00:15:59.520 --> 00:16:01.640 of an interaction, what it means for storytelling. 00:16:01.720 --> 00:16:04.720 I mean, this is the space that you at Runway really work in. 00:16:04.800 --> 00:16:05.440 Yeah. 00:16:05.520 --> 00:16:12.080 So for us, we like to think that it's a little bit different from product 00:16:12.160 --> 00:16:17.360 on a service, where when you are seeking for a product or a service, 00:16:17.440 --> 00:16:22.920 on one side, you have the provider that can tailor 00:16:23.000 --> 00:16:28.920 the experience for you based on your previous purchases, maybe on your likes, 00:16:29.000 --> 00:16:32.080 on things that you have I've searched before. 00:16:32.160 --> 00:16:37.400 And that's great to create hyper personalized experiences for you to maybe 00:16:37.480 --> 00:16:42.880 the next time that I come there, I can get exactly what I want with maybe 00:16:42.960 --> 00:16:46.760 less effort or a hyper personalized experience. 00:16:46.840 --> 00:16:48.160 Whereas in storytelling, 00:16:48.240 --> 00:16:53.600 we believe there might be a different scenario where perhaps hyper 00:16:53.680 --> 00:16:59.280 personalisation doesn't work the same way that it works with a product and service. 00:16:59.360 --> 00:17:00.080 Why? 00:17:00.160 --> 00:17:06.480 Because I don't know how many of you watch succession, the show on HBO. 00:17:06.560 --> 00:17:10.720 Succession happened, the finale happened on a Sunday, right? 00:17:10.800 --> 00:17:16.200 Next day, Monday, we were all talking about what happened on that finale. 00:17:16.280 --> 00:17:18.520 It's likely that we will not be talking 00:17:18.600 --> 00:17:23.440 about personalized stories next day at work because those are so tailored 00:17:23.520 --> 00:17:30.320 to you that maybe that social aspect in storytelling is crucial for creating 00:17:30.400 --> 00:17:36.760 this more emotive connections with storytelling. 00:17:36.840 --> 00:17:39.160 For us, I like to think a lot about 00:17:39.240 --> 00:17:45.840 hyperdiversification where you, instead of have tailored stories for you, 00:17:45.920 --> 00:17:51.440 what would happen if you have right after the finale, you have a different angle 00:17:51.520 --> 00:17:54.040 of the finale that you can also share with others? 00:17:54.120 --> 00:17:58.520 Maybe what if the same story is presented to you by a different character? 00:17:58.520 --> 00:18:00.120 What if maybe the story never ends. 00:18:00.200 --> 00:18:02.280 Maybe have different ending, 00:18:02.360 --> 00:18:05.960 so maybe different ways of seeing different angles of the same story 00:18:06.040 --> 00:18:10.440 that are maybe less of one to one and maybe more one to many. 00:18:10.520 --> 00:18:13.920 Yeah, it's interesting to think about everyone 00:18:13.920 --> 00:18:17.200 has an imagination of what happened after the last scene of a show. 00:18:17.200 --> 00:18:20.520 And so what does that look like and how does that spark new conversations? 00:18:20.600 --> 00:18:23.720 I think the other interesting thing about the work you guys do is 00:18:23.800 --> 00:18:28.920 tools for creators, because I think a lot of times the hype around AI can be very 00:18:29.000 --> 00:18:31.680 much like Well, that's the end of everybody's jobs. 00:18:31.760 --> 00:18:35.560 But there are other ways that people can use these tools 00:18:35.640 --> 00:18:42.800 to tell great stories or to bring an idea to life in a way that maybe previously... 00:18:42.880 --> 00:18:44.320 There's some great case studies on you 00:18:44.320 --> 00:18:47.400 guys' website about the thing that used to take me five hours takes five minutes now. 00:18:47.400 --> 00:18:48.520 But maybe you can talk about how this 00:18:48.520 --> 00:18:50.760 really opens up possibilities for creators as well. 00:18:50.840 --> 00:18:52.440 When we started the company, 00:18:52.520 --> 00:18:56.040 we wanted to create tools for people to tell their stories. 00:18:56.120 --> 00:18:58.080 People who couldn't tell these stories 00:18:58.160 --> 00:19:02.560 before because budget, because of timing, because it takes time. 00:19:02.640 --> 00:19:07.720 There is a learning curve to learn how to edit videos or tell stories. 00:19:07.800 --> 00:19:09.880 But not just for people, not just for creators, 00:19:09.960 --> 00:19:15.800 also for brands and for enterprises, how to package their story. 00:19:15.880 --> 00:19:18.960 But when we created the company, 00:19:18.960 --> 00:19:20.400 we started thinking we come from that background. 00:19:20.480 --> 00:19:23.200 The three of us, the three founders, met at Art School. 00:19:23.280 --> 00:19:27.360 We were doing this things before we started the company. 00:19:27.440 --> 00:19:32.400 And for us, the creator on the loop is crucial. 00:19:32.480 --> 00:19:34.520 It's something that we want to... 00:19:34.600 --> 00:19:37.560 We're creating tools at the end. 00:19:37.640 --> 00:19:43.080 We're enabling creators to tell their stories as they want, 00:19:43.160 --> 00:19:49.280 things that they have on on plan on their heads and they want to communicate. 00:19:49.360 --> 00:19:51.280 So for us, it's 00:19:51.360 --> 00:19:57.960 all about building the necessary tools for them to execute on those ideas. 00:19:58.040 --> 00:20:02.640 And it's never been created in a way that we try to replace the artist. 00:20:02.720 --> 00:20:05.920 We're not thinking on a printer where I 00:20:06.000 --> 00:20:08.440 give you the script, you give me a movie back. 00:20:08.520 --> 00:20:10.760 That's not what we aim to create. 00:20:10.840 --> 00:20:13.600 We're aiming to create tools that can help 00:20:13.680 --> 00:20:18.160 power this ideas and turn them into realities. 00:20:18.240 --> 00:20:19.720 I heard the other day, 00:20:19.800 --> 00:20:28.160 Guillermo Torres saying that the natural state of a movie is not to happen. 00:20:28.240 --> 00:20:31.320 And when they happen, it's actually a miracle. 00:20:31.400 --> 00:20:32.680 And that's true. 00:20:32.760 --> 00:20:37.840 And what we are trying to enable is coming from the three founders come 00:20:37.840 --> 00:20:40.520 from different places, two from Chile and one from Greece. 00:20:40.600 --> 00:20:45.440 We're thinking in the idea of making a movie that last two hours is 00:20:45.520 --> 00:20:49.040 crazy because first, you don't have the crew to do it, 00:20:49.120 --> 00:20:52.080 you don't have the budget, you maybe don't have the ambition 00:20:52.160 --> 00:20:55.840 to create a 2 hours feature film movie because it feels so far. 00:20:55.920 --> 00:21:02.160 But by enabling the tools for you to get a quick access to that and quick feedback 00:21:02.240 --> 00:21:05.680 to visualize those ideas is the thing, the products that we're building. 00:21:05.760 --> 00:21:09.200 I wonder about thinking about personalization, 00:21:09.280 --> 00:21:13.120 thinking about storytelling, thinking about unlocking creativity in new 00:21:13.200 --> 00:21:17.760 ways for people, the way that the socialness of it plays out. 00:21:17.840 --> 00:21:22.160 I wonder, Carly, what you think about that. 00:21:22.240 --> 00:21:24.520 If we're all living in generative worlds 00:21:24.600 --> 00:21:28.960 that are totally tailored to ourselves, what do we talk about at the bus stop? 00:21:29.040 --> 00:21:32.160 What do we talk about at the PTA meeting? 00:21:32.240 --> 00:21:35.960 What are the things that- We were talking earlier and I was like, 00:21:36.040 --> 00:21:41.920 I don't know why I joined the PTA board and I'm a full-time working mother of two. 00:21:42.000 --> 00:21:44.520 I'm in negative hours right now, people. 00:21:44.600 --> 00:21:49.440 But I actually want to first remark on something I think Alejandro was 00:21:49.520 --> 00:21:55.440 explaining is around this idea of centar systems, human plus machines. 00:21:55.520 --> 00:21:57.520 I, as a human, am able to do certain 00:21:57.600 --> 00:22:01.280 things, and the machine, and in itself can do certain things. 00:22:01.360 --> 00:22:03.720 But the human plus the machine can 00:22:03.800 --> 00:22:06.320 ultimately create much more interesting outputs. 00:22:06.400 --> 00:22:08.440 I think that collaboration 00:22:08.520 --> 00:22:14.000 and that dynamic is why I'm also an optimist in terms of why these creative 00:22:14.080 --> 00:22:19.320 tools are so important to empower, to enable, because Sintar systems, 00:22:19.400 --> 00:22:22.760 more often than not, when you also look at the data, 00:22:22.840 --> 00:22:28.840 produce incredible outputs, whether it be in medicine, identifying cancer and so on. 00:22:28.920 --> 00:22:33.960 Then when you ask, Well, what does this look like in terms of social dynamics? 00:22:33.990 --> 00:22:36.040 Let's say I have a personal agent and you 00:22:36.120 --> 00:22:40.240 have an agent, and those agents are all talking to each other. 00:22:40.320 --> 00:22:44.440 I'm going to keep going back to the compute because the whole reason we 00:22:44.520 --> 00:22:48.800 can have generative AI is because of the advancements in compute. 00:22:48.880 --> 00:22:51.280 I just don't want to understate that. 00:22:51.360 --> 00:22:53.200 But I 00:22:53.440 --> 00:22:58.880 do think that it's about augmentation and isn't about replacement, 00:22:58.960 --> 00:23:04.720 and that I hope, is about amplification and deeper connection. 00:23:04.800 --> 00:23:11.200 I also think that you can get more by... 00:23:11.280 --> 00:23:12.800 I mean, we live in an information economy, 00:23:12.880 --> 00:23:17.040 and I can't sort through everything that hits me all the time. 00:23:17.120 --> 00:23:19.680 I want to have deeper connections. I want to be more present. 00:23:19.760 --> 00:23:25.680 I would love when I see my kids that we're talking about our day and I'm not having 00:23:25.760 --> 00:23:29.760 to sort out all of these things across the bazillion of apps that I use. 00:23:29.840 --> 00:23:32.000 Wouldn't it be great to have that? 00:23:32.080 --> 00:23:35.480 I guess I'm an optimist. 00:23:35.560 --> 00:23:37.200 There's maybe some pessimists in the room, 00:23:37.280 --> 00:23:39.920 and if you're pessimist, ask a question at the end 00:23:40.000 --> 00:23:43.440 or meet up for a drink at the bar and have a discussion about that. 00:23:43.520 --> 00:23:48.080 But I think of these agents 00:23:48.160 --> 00:23:51.520 as additional actors, and that's why I go back to that framework 00:23:51.600 --> 00:23:56.200 of relationships and roles and your reaction and how you choose to engage. 00:23:56.280 --> 00:23:59.200 I think this brings us into a thing that I 00:23:59.280 --> 00:24:03.720 want to bring you in on Zyra, which is what we're talking about is not just 00:24:03.800 --> 00:24:06.640 relationship, if we're going to use this word relationship. 00:24:06.720 --> 00:24:09.000 One of the core elements that makes 00:24:09.080 --> 00:24:12.920 something actually a relationship is some modicum of trust. 00:24:13.000 --> 00:24:16.000 At the very least, even if you have a fairly transactional 00:24:16.040 --> 00:24:19.640 relationship, you trust that the transaction will be the same every time. 00:24:19.640 --> 00:24:20.680 You trust that the other side will honor 00:24:20.760 --> 00:24:23.080 the transaction, you'll honor the transaction. 00:24:23.160 --> 00:24:25.440 I want to bring in this idea of trust 00:24:25.520 --> 00:24:31.840 because I think that there is this trust building that needs to be baked 00:24:31.920 --> 00:24:37.160 into how we think about this stuff, in addition to the extremely expensive 00:24:37.240 --> 00:24:40.720 compute costs of bots talking to each other. 00:24:40.800 --> 00:24:45.200 But setting that aside for the moment, what are the building blocks of trust 00:24:45.280 --> 00:24:47.720 based on the work that you've done with algorithms and other things? 00:24:47.720 --> 00:24:49.360 What are the things that brands should be thinking about here? 00:24:49.440 --> 00:24:54.280 Yeah, I wanted to just pick up on, Carly, what you were saying about augmentation, 00:24:54.360 --> 00:25:00.000 because I actually think that we need to learn how to do that. 00:25:00.080 --> 00:25:03.280 Because if I think about it in much higher stakes in what we're talking about, 00:25:03.360 --> 00:25:11.720 we're talking, let's say, to your point, about cancer diagnosis or fighter pilots. 00:25:11.800 --> 00:25:19.000 When I'm looking at that AI, as a human, sometimes I might under trust 00:25:19.080 --> 00:25:22.440 the technology or I might over trust the technology. 00:25:22.520 --> 00:25:28.280 And we actually need to figure out what is the right augmentation look like. 00:25:28.360 --> 00:25:31.360 And so that will take a long time to figure that out. 00:25:31.440 --> 00:25:33.760 I don't think we're there yet. 00:25:33.840 --> 00:25:38.000 We have to figure out, first of all, what are we building? 00:25:38.080 --> 00:25:41.920 How do we build that relationship with human and machine? 00:25:42.000 --> 00:25:46.520 And then there's a lot of work that's been 00:25:46.600 --> 00:25:51.880 done on how do you actually add more friction into the relationship. 00:25:51.960 --> 00:25:56.560 And the reason for that is you want the human to be able to take a step back, 00:25:56.640 --> 00:26:00.000 stop and say, Is this actually what I should be doing? 00:26:00.080 --> 00:26:02.640 And Joe, you were saying that earlier with the designers. 00:26:02.640 --> 00:26:05.160 Exactly. Should we actually be doing that? 00:26:05.240 --> 00:26:07.800 So you need lots of moments of extra 00:26:07.880 --> 00:26:12.320 friction along the way to make people pause and say, Do I trust this? 00:26:12.400 --> 00:26:14.480 I was going to say not just designers, 00:26:14.560 --> 00:26:19.160 but like, government organizations like Data and Trust Alliance. 00:26:19.240 --> 00:26:25.720 I think this is why the AI Act that came out in European Union, 00:26:25.800 --> 00:26:31.160 it has a whole section on emotional AI, and it puts most of it in the unacceptable 00:26:31.240 --> 00:26:35.880 risk, in the high risk category, except in a few specific use cases 00:26:35.960 --> 00:26:40.080 for some of the reasons you just mentioned, because these are very high 00:26:40.160 --> 00:26:44.720 stakes life or death decisions that have to hold up in court. 00:26:44.800 --> 00:26:47.760 They have to hold up in front of Congress, things like that. 00:26:47.840 --> 00:26:51.120 Can we double click, though, on the trust element? 00:26:51.200 --> 00:26:55.560 Because when you think about when you build trust with a person, 00:26:55.640 --> 00:27:01.880 it's off of consistent behavior and meeting value system to some extent. 00:27:01.880 --> 00:27:04.040 I actually wanted to ask you this before we came down here. 00:27:04.120 --> 00:27:06.520 I was like, how do you define trust? 00:27:06.600 --> 00:27:11.480 Because for me, it needs to be a shared value system. 00:27:11.480 --> 00:27:13.560 And I know you're not going to disappoint me. 00:27:13.640 --> 00:27:15.800 And so you have my trust. 00:27:15.880 --> 00:27:18.520 And you don't get it, you earn it. 00:27:18.600 --> 00:27:20.080 And in order to earn something, 00:27:20.160 --> 00:27:22.840 that means that you're having an engagement over time. 00:27:22.920 --> 00:27:25.320 But you need to adopt the AI. 00:27:25.400 --> 00:27:27.520 You shouldn't just blindly take it. Absolutely. 00:27:27.600 --> 00:27:30.080 I think that's what you were starting at. 00:27:30.160 --> 00:27:31.840 That's a fabulous point. 00:27:31.920 --> 00:27:36.840 The Data and Trust Alliance is an alliance of about 25 different companies, 00:27:36.920 --> 00:27:39.720 each of which are very influential in their ecosystem. 00:27:39.800 --> 00:27:43.040 So huge companies think Nike, CVS, et cetera. 00:27:43.120 --> 00:27:44.240 And to your point, 00:27:44.320 --> 00:27:48.600 when we talk about trust, it is all about, these guys are making 00:27:48.680 --> 00:27:54.840 phenomenal tech, but do I, as a company or as an organization, trust it? 00:27:54.840 --> 00:27:56.680 That means I need to have my own value 00:27:56.760 --> 00:28:01.840 system, and I need to be able to understand whether that tech fits it. 00:28:01.920 --> 00:28:03.840 And so we're now just at the beginning 00:28:03.920 --> 00:28:06.680 of being able to understand, well, if I look at a... 00:28:06.760 --> 00:28:08.560 I'm not going to be building a foundation 00:28:08.640 --> 00:28:13.720 model in my company because these things cost $500 million to build. 00:28:13.720 --> 00:28:15.560 None of these companies are going to be 00:28:15.640 --> 00:28:19.680 making their own models, but they are going to be using third-party models. 00:28:19.760 --> 00:28:23.600 And so then the question is, which third-party models should I use? 00:28:23.680 --> 00:28:25.720 Which one do I trust? 00:28:25.800 --> 00:28:29.280 Which one aligns with my values, my policies? 00:28:29.360 --> 00:28:31.040 How do I even to get that out? 00:28:31.120 --> 00:28:35.600 We're actually quite early in being able to figure that out. 00:28:36.200 --> 00:28:38.480 So that's how we do it. 00:28:38.560 --> 00:28:39.720 And here's exactly... 00:28:39.800 --> 00:28:43.880 You just gave me my first perfect segue from a panelist today. 00:28:43.880 --> 00:28:45.280 Because, Joe, I wanted to bring you in on this? 00:28:45.360 --> 00:28:46.760 Because I feel like every new round 00:28:46.840 --> 00:28:50.160 of technology, one of the big questions is, do we build? 00:28:50.160 --> 00:28:51.440 Do we buy? Do we partner? 00:28:51.520 --> 00:28:53.120 I think particularly when you're thinking 00:28:53.200 --> 00:29:00.520 about a portfolio of brands, each with their own contexts, customers, 00:29:00.600 --> 00:29:05.080 and stakes, how do you think about whether to build, 00:29:05.160 --> 00:29:07.920 buy, partner, how to approach this problem? 00:29:08.000 --> 00:29:10.040 Initially, we worked with partnerships, 00:29:10.120 --> 00:29:13.520 so licensing all the top foundational models is where we started. 00:29:13.600 --> 00:29:18.000 But I think to Carly's point, eventually it gets down to compute. 00:29:18.080 --> 00:29:23.800 And for a corporation of Intuit's size, it gets very expensive very quickly. 00:29:23.880 --> 00:29:28.880 And at some point, we have to look at what does it take to build our own foundational 00:29:28.960 --> 00:29:34.000 model, Maybe one that specializes in the domains that our customers expect, 00:29:34.080 --> 00:29:40.040 ones that we can validate the authenticity with. 00:29:40.120 --> 00:29:42.920 So right 00:29:43.800 --> 00:29:47.520 now, we've started with essentially a partnership model, 00:29:47.600 --> 00:29:52.280 but I think a lot of companies will be moving into build their own models. 00:29:52.360 --> 00:29:56.800 I think Bloomberg here has their own model that they've been building. 00:29:56.880 --> 00:30:01.600 And so I'd imagine that'll be coming with lots of companies. 00:30:01.680 --> 00:30:02.560 I think it's going to be a really 00:30:02.640 --> 00:30:08.200 interesting ecosystem where some people are going to be 00:30:08.280 --> 00:30:12.360 licensing a model and then training it on their own data versus just building 00:30:12.440 --> 00:30:16.280 something from scratch, which is like, why? 00:30:16.640 --> 00:30:18.000 But I ain't doing that. 00:30:18.000 --> 00:30:19.240 But other companies might. 00:30:19.320 --> 00:30:29.240 What are the challenges there for brands that are just starting to dive 00:30:29.240 --> 00:30:32.480 into this space Okay, so maybe, Maz, you've started to think about this as you 00:30:32.480 --> 00:30:35.040 think about the clients you work with wanting to create these more emotive 00:30:35.120 --> 00:30:38.440 experiences, but what's the right way to get started? 00:30:38.520 --> 00:30:40.480 Yeah. I mean, everything that you folks have 00:30:40.560 --> 00:30:44.880 been saying around relationship building is really fascinating because number one, 00:30:44.910 --> 00:30:48.960 the brand, the business a little bit needs to figure out their relationship to AI. 00:30:49.040 --> 00:30:55.240 What is it about our experience that we think AI could usefully augment? 00:30:55.320 --> 00:30:58.960 And how do we define the role of our 00:30:58.960 --> 00:31:02.640 people versus the role of our agents and the systems that are more automated. 00:31:02.720 --> 00:31:10.400 There is that aspect of really a little bit of education and bit by bit adoption. 00:31:10.480 --> 00:31:12.360 Experimentation is critical. 00:31:12.440 --> 00:31:14.600 Something that we really advocate for as 00:31:14.680 --> 00:31:19.520 a design studio is make, test, learn as a concept, very familiar to a lot 00:31:19.600 --> 00:31:23.480 of you, I'm sure, but extra essential when you're considering 00:31:23.560 --> 00:31:27.280 the stakes of the kinds of technologies that we're talking about here, 00:31:27.360 --> 00:31:31.360 especially as it relates to engaging engaging with customers. 00:31:31.440 --> 00:31:37.680 So moving on to the how do we approach customers building relationships with our 00:31:37.760 --> 00:31:42.160 tools, with our systems that are more AI-powered or even AI-overseen. 00:31:42.240 --> 00:31:44.560 I think it's the same way that you build a relationship with a human. 00:31:44.640 --> 00:31:48.800 It's the same way that when you're first dating somebody, you're not going to go up 00:31:48.800 --> 00:31:51.400 to a cabin in the woods with them and take their tax advice. 00:31:51.480 --> 00:31:56.760 You're maybe going to say, Hey, I'd like to get a little bit of a tailored 00:31:56.810 --> 00:31:59.760 to my knowledge level guide to a dummy's guide to investing, 00:31:59.840 --> 00:32:05.280 or maybe I need to understand the basics of saving for a house or what have you. 00:32:05.330 --> 00:32:08.480 You're going to start small and you're going to build up and you 00:32:08.560 --> 00:32:14.000 need to trust but verify your way to the level 00:32:14.080 --> 00:32:18.440 of comfort that you want to arrive at with regards to your relationship with AI. 00:32:18.520 --> 00:32:22.200 I think for both brands, they need to build their comfort and trial 00:32:22.280 --> 00:32:27.760 and learn what is an acceptable error rate or what have you. 00:32:27.840 --> 00:32:30.080 What caveats do we put in place? 00:32:30.130 --> 00:32:31.880 What watermarks do we put in place? 00:32:31.960 --> 00:32:33.400 That is all essential. 00:32:33.480 --> 00:32:37.080 But then on the consumer side, on the user-facing side, 00:32:37.160 --> 00:32:40.760 that relationship building needs to take place incrementally and iteratively, too. 00:32:40.840 --> 00:32:42.520 I wonder, Alejandro, 00:32:42.600 --> 00:32:47.360 what you have seen at Runway with your customers and how they onboard to the use 00:32:47.440 --> 00:32:51.920 of these tools and adopt them and feel their way through how they want to use 00:32:52.000 --> 00:32:55.520 machine learning and AI to create and tell stories. 00:32:55.600 --> 00:33:00.920 Yeah, we see a lot of different use cases Yes, 00:33:01.000 --> 00:33:04.880 from a brand perspective or enterprise for narrative planning, 00:33:04.960 --> 00:33:12.720 for conceptualizing some of the ideas for either a campaign or for a concept. 00:33:12.720 --> 00:33:14.880 They're exploring. There's a lot of what Matt is saying, 00:33:14.960 --> 00:33:21.800 a lot of tinkering still of how can I apply this to the workflows. 00:33:22.360 --> 00:33:31.560 Back to the conversation in terms of how can I use this on my a company in terms 00:33:31.640 --> 00:33:37.720 of outsourcing a model or maybe buying or using one. 00:33:37.800 --> 00:33:40.440 In our case, we have seen that for us it's 00:33:40.520 --> 00:33:46.560 crucial to invest in creating our own models because as we're working 00:33:46.640 --> 00:33:50.640 with creators, we know that control is fundamental. 00:33:50.720 --> 00:33:55.800 As much as we can provide control to what 00:33:55.880 --> 00:33:58.800 they're seeking for is it provides a better tool. 00:33:58.880 --> 00:34:01.120 So in order to When you have full control 00:34:01.200 --> 00:34:08.240 of the outputs, you need to understand how the model works, how the model is created, 00:34:08.240 --> 00:34:11.400 the architecture of the model is, how it's deployed in the model, 00:34:11.480 --> 00:34:14.840 and how you build applications on top of the model. 00:34:14.920 --> 00:34:18.280 So for us, in that case, 00:34:18.360 --> 00:34:24.280 building, knowing how the full stack works, helps us to provide that grade 00:34:24.360 --> 00:34:29.280 of control that we can give to our customers and users. 00:34:29.360 --> 00:34:34.320 Then When some users have some ideas of maybe what we're talking about, 00:34:34.400 --> 00:34:39.080 context, where maybe I have some context that wasn't not part of the training data 00:34:39.160 --> 00:34:43.360 set of what my foundation model was, then it's when maybe it comes to a second 00:34:43.440 --> 00:34:48.000 layer of fine-tuning on top of that model, then maybe I just need one extra layer 00:34:48.080 --> 00:34:53.520 on top of that to still remain all grade of control that I can get from your base 00:34:53.600 --> 00:34:57.000 model plus your applications, but I can add this second layer on top 00:34:57.080 --> 00:35:00.720 where the context that I need to provide fit my needs. 00:35:00.800 --> 00:35:01.840 Yeah, go ahead, go. 00:35:01.920 --> 00:35:04.640 I'm just a visual person, so I was always listening to people. 00:35:04.720 --> 00:35:07.840 It's like I have this diagram in my head. 00:35:08.120 --> 00:35:14.640 It's like when we're talking about the relationship with AI is that you have 00:35:14.720 --> 00:35:19.600 You have creative and proactive, and you have left brain and right brain. 00:35:19.680 --> 00:35:24.520 It's like your analytical use cases and your creative use cases. 00:35:24.600 --> 00:35:26.600 You have very proactive engagement, 00:35:26.680 --> 00:35:31.520 and you have very reactive in the sense that I invoke access to certain things. 00:35:31.600 --> 00:35:35.640 And I think that that two by two 00:35:35.720 --> 00:35:41.320 and thinking about use cases within that for brands is maybe a helpful way. 00:35:41.400 --> 00:35:46.600 And then marry that with when you're asking about this whole model situation. 00:35:46.680 --> 00:35:49.080 Personally, I think there's a strong case 00:35:49.160 --> 00:35:54.480 that there are going to be a few winners with foundational models. 00:35:54.560 --> 00:35:56.800 And I think there's a strong argument 00:35:56.880 --> 00:36:03.440 for fine-tuning, bring your own data, and inference services because 00:36:03.520 --> 00:36:09.840 I just think the cost of entry, to your point, is really high. 00:36:09.920 --> 00:36:12.840 Also the number of parameters that these 00:36:12.920 --> 00:36:16.280 larger models have are just so significant. 00:36:16.360 --> 00:36:18.680 If we were here talking about narrow AI 00:36:18.760 --> 00:36:21.400 use cases, maybe it would be a different answer. 00:36:21.480 --> 00:36:29.400 But with these LLMs and future multi-model models, I think the argument 00:36:29.480 --> 00:36:32.160 for partnering doing is going to be really strong. 00:36:32.240 --> 00:36:33.720 Yeah. 00:36:33.840 --> 00:36:35.280 Then I want to bring it back to the work 00:36:35.360 --> 00:36:40.320 you're doing at DATA+ Trust, Zyra, because I think the pressure is very much 00:36:40.400 --> 00:36:46.360 on with this technology in a way that I think it wasn't before to have regulators 00:36:46.360 --> 00:36:50.160 feel like they got their arms around it, to have large companies who could be held 00:36:50.240 --> 00:36:54.280 liable for the decisions that the software makes, to get their arms around it. 00:36:54.360 --> 00:36:58.160 I wonder if there are things that you've learned, your team has learned 00:36:58.240 --> 00:37:02.320 from looking at algorithmic bias and other things that might be applicable here when 00:37:02.320 --> 00:37:05.720 you're vetting partners, when a company is considering which partners to go with. 00:37:05.800 --> 00:37:09.760 When we first started looking at this, 00:37:09.840 --> 00:37:18.040 we realized that actually what we need to start assessing is, do I trust the data? 00:37:18.120 --> 00:37:20.480 Do I trust the models? 00:37:20.560 --> 00:37:24.720 And do I trust the people and the processes that sit around them? 00:37:24.800 --> 00:37:26.400 So that might be at a vendor, 00:37:26.480 --> 00:37:31.480 but also in our company, have we set it up in the right way. 00:37:31.560 --> 00:37:37.080 And so when we first did our very first project, 00:37:37.160 --> 00:37:42.920 we went into the world of HR because it is a very high risk application of AI. 00:37:43.000 --> 00:37:46.160 And we said, okay, are you guys using AI in HR? 00:37:46.240 --> 00:37:48.600 Yes. Are you planning on using it more? 00:37:48.680 --> 00:37:49.840 Yes. 00:37:49.920 --> 00:37:52.200 Are you building these things in-house? 00:37:52.280 --> 00:37:53.400 Are you buying them? 00:37:53.400 --> 00:37:54.440 Well, we're buying them. 00:37:54.520 --> 00:37:58.560 What are you concerned about in the world of HR bias? 00:37:58.640 --> 00:38:03.240 Okay, well, are you asking any of these companies their approach to bias? 00:38:03.320 --> 00:38:05.520 And the answer was no. 00:38:05.600 --> 00:38:09.480 So then we had to build a whole vendor evaluation tool so that if you're bringing 00:38:09.560 --> 00:38:13.840 AI into your world, you're not just bringing it on the basis 00:38:13.840 --> 00:38:16.920 of like, Okay, is this the right cost or performance or whatever? 00:38:17.000 --> 00:38:18.560 But I'm asking you, 00:38:18.640 --> 00:38:24.280 do you know how to detect or mitigate or monitor algorithmic bias? 00:38:24.360 --> 00:38:28.720 And so now we're finding that actually somehow the Data and Trust Alliance is 00:38:28.800 --> 00:38:33.880 turning into Vendor Evaluation Central because they're all... 00:38:33.960 --> 00:38:39.320 To your point, it's going to be a handful of folks that are doing foundation models. 00:38:39.400 --> 00:38:41.400 And then if we're partnering with them 00:38:41.480 --> 00:38:46.880 to finetune it, you've actually already trained the model with your data. 00:38:46.960 --> 00:38:48.800 I need to know what that was, 00:38:48.880 --> 00:38:54.760 because even if I'm giving you my data to finetune it, it was already trained. 00:38:54.840 --> 00:38:56.640 So I need to know. 00:38:56.640 --> 00:39:00.000 And one of the most important questions these companies should be asking actually 00:39:00.080 --> 00:39:04.520 is, what are the unintended consequences of me taking this on? 00:39:04.600 --> 00:39:05.240 No. 00:39:05.320 --> 00:39:09.640 If you're not actually asking that question to yourself, 00:39:10.200 --> 00:39:14.760 regulation or not, you might find yourself in a very sticky situation. 00:39:14.840 --> 00:39:15.560 Yeah. 00:39:15.640 --> 00:39:18.440 I wonder if you have any reflections on that, Joe, for the kinds of products 00:39:18.520 --> 00:39:22.880 and people that you're trying to serve, thinking about. 00:39:22.960 --> 00:39:25.040 Yeah. The thing that's immediately calling 00:39:25.120 --> 00:39:29.960 to mind is Mayor Adams chatbot that's advising you on the employment status 00:39:30.040 --> 00:39:36.120 of your bodega cat and giving people some advice to break the law. 00:39:36.200 --> 00:39:39.720 But maybe a more serious actor maybe has some. 00:39:39.720 --> 00:39:39.800 Yeah. 00:39:39.880 --> 00:39:45.080 Is there any MailChimp folks in the house tonight, for any chance? 00:39:45.160 --> 00:39:46.360 No. 00:39:47.160 --> 00:39:52.320 The Mailchimp use cases are really interesting, but there's some of the most, 00:39:52.400 --> 00:39:56.280 I think, obvious ones for generative AI because 00:39:56.360 --> 00:40:01.120 managing email campaigns and all the content that gets generated, fairly 00:40:01.520 --> 00:40:04.520 low risk other than representation of your brand. 00:40:04.600 --> 00:40:07.400 That's very different than some other use 00:40:07.480 --> 00:40:12.640 cases like lending money, where bias is a huge issue. 00:40:12.720 --> 00:40:13.640 I don't know if you remember a couple 00:40:13.720 --> 00:40:18.680 of years ago, I think Apple got busted when they were doing payments 00:40:18.760 --> 00:40:22.600 acceptance, where they found out that the data they trained on was biased, 00:40:22.640 --> 00:40:25.440 and they were eliminating all these people from payments. 00:40:25.520 --> 00:40:29.000 So I think that's a case where 00:40:29.080 --> 00:40:34.840 the data that you to use has to be completely unbiased, run by an alliance or 00:40:34.920 --> 00:40:41.440 a third party, validated just because of the stakes being so high. 00:40:41.520 --> 00:40:43.520 I think we're talking about data. 00:40:43.520 --> 00:40:44.920 There's also features, right? 00:40:45.000 --> 00:40:47.600 That makes sense between different data sets. 00:40:47.680 --> 00:40:51.560 One of the things... I used to work at J. 00:40:51.560 --> 00:40:55.920 P. Morgan, and we built an ensemble model 00:40:56.000 --> 00:41:01.640 and a way to help with building trust and 00:41:01.960 --> 00:41:06.360 making it feel more approachable to use 00:41:06.360 --> 00:41:09.440 the output because it was probabilistic, was to include confidence intervals 00:41:09.520 --> 00:41:12.680 because there was a lot of noise, for example, in the data. 00:41:12.760 --> 00:41:16.360 I think that you're never going to have a perfect data set. 00:41:16.440 --> 00:41:20.600 I think you can do enrichment to make it more consistent, more complete. 00:41:20.680 --> 00:41:22.520 There's underrepresentation in certain 00:41:22.600 --> 00:41:26.200 populations, which makes information inherently biased. 00:41:26.280 --> 00:41:28.480 Being able to identify that. 00:41:28.560 --> 00:41:35.960 But on the of data and trust, I do think that where you can design 00:41:35.960 --> 00:41:39.240 explainability, things like confidence intervals when you're talking about 00:41:39.320 --> 00:41:45.040 financial data is an interesting approach with trust, because it's in a relationship 00:41:45.120 --> 00:41:48.520 you share something about yourself to help build that trust. 00:41:48.600 --> 00:41:54.240 I think that's maybe a second element outside of the consistency of engagement. 00:41:54.320 --> 00:41:56.600 Yeah. Go ahead, Zara. 00:41:56.680 --> 00:41:57.800 Are you interested in that? I'm sorry. 00:41:57.880 --> 00:42:00.880 I think one of the hallucinating you having thoughts. 00:42:00.960 --> 00:42:03.320 I'm like an AI. 00:42:03.400 --> 00:42:09.760 One of the things I wonder about is we've been now talking chiefly about, 00:42:09.760 --> 00:42:11.680 from the brand perspective, the business perspective, 00:42:11.680 --> 00:42:14.280 the technologist perspective, even a little bit the regulator's 00:42:14.360 --> 00:42:18.560 perspective, I wonder about the going back 00:42:18.640 --> 00:42:23.560 to the customer experience and how much do we really need to know? 00:42:23.640 --> 00:42:26.360 I know that there's been headlines of late 00:42:26.440 --> 00:42:29.520 about what happens when you click on that little info button and it tells 00:42:29.600 --> 00:42:33.480 you your prompt was transformed, and you're like, That's not what I meant. 00:42:33.560 --> 00:42:36.160 I'm curious about how you're thinking 00:42:36.240 --> 00:42:40.400 about that from the customer side of the trust equation. 00:42:40.480 --> 00:42:42.800 There's what you can do as designers 00:42:42.880 --> 00:42:46.720 and developers and brands, but what is the other side of that equation? 00:42:46.800 --> 00:42:48.600 How do we avoid emotive AI becoming 00:42:48.680 --> 00:42:52.760 manipulative AI or people just by default not trusting it? 00:42:52.840 --> 00:42:57.200 I'll open up to anybody who wants to take a stab at that. 00:42:57.280 --> 00:43:00.440 I'll open it up with a It's a debate 00:43:00.520 --> 00:43:03.720 that's ongoing, and I think you brought it up around labeling. 00:43:03.800 --> 00:43:08.760 You said watermarked, and then you talked about labeling. 00:43:08.840 --> 00:43:13.360 Some people think that customers need everything labeled to know if it's AI. 00:43:13.440 --> 00:43:15.360 They're like, I need to know. 00:43:15.440 --> 00:43:19.600 Other people are like, Well, okay, if I start to label everything, 00:43:19.680 --> 00:43:25.480 the things that don't get labeled, is that inherently trustful content? 00:43:25.560 --> 00:43:31.280 People can be just as deceitful and deceptive as technology. 00:43:31.360 --> 00:43:33.360 I mean, I think we've seen that, right? 00:43:33.440 --> 00:43:37.280 I think that there is work to be done 00:43:37.360 --> 00:43:42.120 on what does it mean for customers engaging with this type of content. 00:43:42.200 --> 00:43:46.440 But if we also go back to the discussion of emotive AI, it's like 00:43:46.520 --> 00:43:51.400 if AI is having you experience a feeling, I mean, you were talking about 00:43:51.400 --> 00:43:54.120 storytelling, and every good story should make you feel something. 00:43:54.200 --> 00:43:57.040 It has a beginning, a middle, an end. 00:43:57.040 --> 00:44:00.040 There's a climax, there's a protagonist, and something What happens? 00:44:00.120 --> 00:44:02.760 And I think that happens in product experiences. 00:44:02.840 --> 00:44:04.560 You have the Nux. 00:44:04.640 --> 00:44:05.960 What do you do with the Nux? 00:44:06.040 --> 00:44:08.040 How are you setting expectations? 00:44:08.040 --> 00:44:09.040 That's the beginning. 00:44:09.120 --> 00:44:10.840 The middle is your engagement. 00:44:10.840 --> 00:44:11.960 What's happening there? 00:44:12.040 --> 00:44:13.200 And how is it learning? 00:44:13.200 --> 00:44:14.520 Is it a competent listener? 00:44:14.600 --> 00:44:17.560 Are you providing affordances to get feedback? 00:44:17.640 --> 00:44:18.520 What's the end? 00:44:18.600 --> 00:44:22.120 Did I achieve my output or the post-conditions met? 00:44:22.120 --> 00:44:23.320 And what is it doing with that? 00:44:23.400 --> 00:44:24.880 And is it evolving with me? 00:44:24.960 --> 00:44:27.640 So I guess I think that we're still trying 00:44:27.720 --> 00:44:32.960 to figure that out because Regulations label everything. 00:44:33.040 --> 00:44:35.800 But what happens when something then isn't? 00:44:35.880 --> 00:44:37.080 Is it automatically authentic? 00:44:37.160 --> 00:44:39.480 No, right? Do you have a thought? 00:44:39.480 --> 00:44:41.760 We were starting to talk a little bit about-Yeah, we were talking about this. 00:44:41.840 --> 00:44:44.960 I think that as unsexy as it is, 00:44:45.040 --> 00:44:49.400 this is where things like standards and certifications come in. 00:44:49.480 --> 00:44:54.920 There's this light in our room. And here, 00:44:55.000 --> 00:44:59.560 the amount of time it took and the certifications it took to make 00:44:59.640 --> 00:45:06.080 sure that that light is hanging there and we're all cool with it was insane. 00:45:06.160 --> 00:45:09.840 And it just shows you that we're so early in this journey. 00:45:09.920 --> 00:45:13.080 We need standards, we need certifications. 00:45:13.160 --> 00:45:15.600 And at some point, Yes, 00:45:15.600 --> 00:45:17.840 there should be a lot of labeling, but at some point we're going to forget 00:45:17.860 --> 00:45:21.160 all the labeling because we're going to start to trust all of that. 00:45:21.240 --> 00:45:22.800 But we actually need to go through 00:45:22.880 --> 00:45:26.960 the process of figuring out what needs to be labeled, 00:45:27.040 --> 00:45:30.640 what needs to be certified, what are the standards we need, et et cetera. 00:45:30.720 --> 00:45:32.960 And then I think we'll be on our way. 00:45:33.040 --> 00:45:35.520 I was just going to add, I think we're in this weird phase two 00:45:35.600 --> 00:45:40.160 from an interaction perspective where we're dealing with this legacy of digital 00:45:40.240 --> 00:45:44.800 assistants and chat bots where we've all lost a lot of trust in them. 00:45:44.880 --> 00:45:47.120 Over the years. And yet the primary use cases 00:45:47.200 --> 00:45:51.800 of interaction we see for the LLMs is basically like a chat interface. 00:45:51.880 --> 00:45:56.520 So we're trying to build back up that trust that we lost for decades. 00:45:56.600 --> 00:46:01.520 I think some of the most interesting interactions maybe have yet to come onto 00:46:01.600 --> 00:46:05.640 the market or have yet for us to adapt to them. 00:46:05.720 --> 00:46:11.560 If you think about language being like 90 % nonverbal, for instance. 00:46:11.640 --> 00:46:13.760 If you think about all the natural UI 00:46:13.840 --> 00:46:18.840 technologies that had been their nascent stages and never got widely adopted. 00:46:18.920 --> 00:46:22.840 And now that we have Gen AI, I think it's just going to be 00:46:22.870 --> 00:46:25.520 a leap forward in terms of what the possibilities are. 00:46:25.600 --> 00:46:33.160 And hopefully we move the interaction out of this chat-based prompt-based box, Luke? 00:46:33.240 --> 00:46:34.280 Yeah, totally. 00:46:34.360 --> 00:46:37.040 I think the conversation around 00:46:37.120 --> 00:46:41.080 factual content recommendations, what have you, is super important. 00:46:41.160 --> 00:46:43.360 But a whole different frontier 00:46:43.440 --> 00:46:49.120 for Emotive AI is really in generative interfaces and the way that we engage 00:46:49.200 --> 00:46:53.600 with experiences becoming more personalized with the help of generative 00:46:53.680 --> 00:46:57.920 AI tools and something that Alejandro can possibly speak to as well. 00:46:58.000 --> 00:47:02.560 But basically the idea that maybe the content of the interface stays 00:47:02.640 --> 00:47:06.080 nice and regulated, nice and safe, nice and same, same for everybody. 00:47:06.160 --> 00:47:10.840 But the way that it's presented, depending on your knowledge level, 00:47:10.920 --> 00:47:14.120 your particular mood, whether you're more of a visual person, 00:47:14.200 --> 00:47:20.600 you prefer to engage in things more in two by twos or in text or in ones and zeros. 00:47:20.680 --> 00:47:25.160 Basically, that is a level of personalization. 00:47:25.240 --> 00:47:28.160 I totally just snored it. 00:47:28.240 --> 00:47:30.240 Just give I'll take care of it. I'll take care of it. 00:47:30.240 --> 00:47:30.480 I'll take care of it. I'll take care of it. 00:47:30.480 --> 00:47:32.520 I'll take care of it. I think in two by twos. 00:47:32.600 --> 00:47:40.840 But yeah, I think that's a very safe but incredibly potentially fruitful form 00:47:40.920 --> 00:47:44.840 of personalization that keeps us on the straight and narrow as far as 00:47:44.920 --> 00:47:49.200 content, but really changes the medium in a way that is just that much more 00:47:49.280 --> 00:47:51.720 compelling, that much more accessible for you. 00:47:51.800 --> 00:47:53.800 And then hopefully, you as a user, 00:47:53.800 --> 00:47:57.440 if you're getting information presented to you in a way that gels with the way 00:47:57.440 --> 00:48:01.480 that you like to pay attention and then lean in, you're actually going to pay more 00:48:01.480 --> 00:48:04.800 attention to that information, and you're going to take that information 00:48:04.880 --> 00:48:08.640 in and hopefully act in a way that's more responsible and more trustful. 00:48:08.720 --> 00:48:11.080 Yeah. Go ahead, Ali. 00:48:11.160 --> 00:48:14.600 I would like to add something to that. 00:48:14.680 --> 00:48:20.040 Yes, I guess reemphasizing on the idea that we are still very early. 00:48:20.240 --> 00:48:26.920 At Runway, we like to go back to previous technological advances 00:48:27.000 --> 00:48:36.840 a lot, and from the the creation of two paintings to the camera to filmmaking. 00:48:37.160 --> 00:48:39.640 It always happened that we tried 00:48:39.720 --> 00:48:44.680 to implement these new technology advances to what we already know how it works. 00:48:44.760 --> 00:48:49.080 We use the camera to take photos of people 00:48:49.160 --> 00:48:55.640 like we used to paint them or we used to record or film like plays. 00:48:55.720 --> 00:48:57.760 But then a few years in, 00:48:57.840 --> 00:49:02.960 we created cinema or we use the camera to create different art. 00:49:03.040 --> 00:49:09.800 Or then, Van Gogh was born because 00:49:10.800 --> 00:49:14.600 the painting on tubes existed. 00:49:15.240 --> 00:49:17.200 I guess we're still in that phase today. 00:49:17.280 --> 00:49:20.360 We're trying to understand these systems 00:49:20.440 --> 00:49:25.280 in a way that we know how we have been working in the past. 00:49:25.360 --> 00:49:27.520 We are trying to seek for personalized 00:49:27.600 --> 00:49:33.600 experiences on top of the that we have already built 10 years ago. 00:49:33.680 --> 00:49:37.880 We have systems built, and now we're trying to add a chat base on top of that. 00:49:37.960 --> 00:49:39.280 Or we're trying to maybe think, 00:49:39.360 --> 00:49:43.600 maybe this UI doesn't need to be fixed and maybe it can be generative and maybe 00:49:43.680 --> 00:49:49.440 it can the response to your emotions or to your previous. 00:49:49.520 --> 00:49:52.240 But the question to me is, doesn't it need to be like that? 00:49:52.320 --> 00:49:56.880 This is only a response to what we're seeing today based on what we already 00:49:56.880 --> 00:49:58.840 know, or maybe it can be completely different. 00:49:58.920 --> 00:50:00.680 We have talked about agents here. 00:50:00.760 --> 00:50:04.880 And maybe the whole concept is not that you're going to go and interact 00:50:04.960 --> 00:50:07.720 with an agent every time that you talk to a brand. 00:50:07.800 --> 00:50:10.240 Maybe you talk to one agent that there is 00:50:10.320 --> 00:50:14.680 your own agent and your agent will figure out anything, things for you. 00:50:14.760 --> 00:50:18.280 And Maybe the only experience that you need to build is 00:50:18.360 --> 00:50:22.200 within your agent, and then you customize it the way that you create, 00:50:22.280 --> 00:50:26.480 and then your agent figured out how you interact with different brands. 00:50:26.560 --> 00:50:29.000 This is, of course, like a speculation. 00:50:29.080 --> 00:50:32.800 We don't know how it's going to play, but what I'm trying to say is in 00:50:32.880 --> 00:50:37.080 most of the past experiences where we have seen 00:50:37.160 --> 00:50:42.440 technological transformations like this, we first tried to plug it in into what we 00:50:42.520 --> 00:50:46.120 already know, but it ended up being It's really different things. 00:50:46.200 --> 00:50:47.200 Yeah. 00:50:47.280 --> 00:50:49.680 So I'm mindful that we've only got a couple of minutes left. 00:50:49.760 --> 00:50:54.080 And so do we want to see if anyone has a question or two before we wrap up? 00:50:54.160 --> 00:50:58.240 And I can pop. No. 00:50:58.240 --> 00:50:59.520 Max. Hi. 00:50:59.600 --> 00:51:04.280 I have a generative AI consultancy for CPG brands and beauty brands. 00:51:04.360 --> 00:51:08.200 The biggest question that's really coming out from a lot of these brands is 00:51:08.280 --> 00:51:12.160 in regards to labeling, what will our customers actually think? 00:51:12.240 --> 00:51:15.280 The EU has obviously released legislation 00:51:15.360 --> 00:51:17.840 that generative AI content must be labeled. 00:51:17.840 --> 00:51:21.000 It's going to be interesting to see what happens here in the States. 00:51:21.080 --> 00:51:24.880 My question is really around FTC and advertising. 00:51:24.880 --> 00:51:26.920 What is going to be the consequences for these brands? 00:51:26.920 --> 00:51:30.440 Is it going to be considered false advertising if the content they're using 00:51:30.520 --> 00:51:33.240 to advertise their products is with generative AI? 00:51:33.320 --> 00:51:34.400 Yeah, I don't know. 00:51:34.480 --> 00:51:36.440 Do you want to take this one? 00:51:36.760 --> 00:51:38.880 I mean, that's a tough question because 00:51:38.960 --> 00:51:45.120 you're talking about who's going to be to blame for the content. 00:51:45.200 --> 00:51:47.000 I mean, that's why I think 00:51:47.080 --> 00:51:51.320 the labeling and showing if it's first-party, third-party. 00:51:51.400 --> 00:51:55.160 I mean, the challenge is that there's so many creator tools now. 00:51:55.240 --> 00:51:57.200 You talked about Loving Midjourney. 00:51:57.280 --> 00:52:03.920 It's like, I don't think that they do water their marks on their images. 00:52:04.000 --> 00:52:08.360 Yeah, right? Then you're at the mercy, right? 00:52:08.440 --> 00:52:13.880 If somebody goes in, they create that to then the distributors. 00:52:13.960 --> 00:52:16.280 I know I I said, I wasn't going to... 00:52:16.360 --> 00:52:19.040 This is my personal point of view. 00:52:19.120 --> 00:52:24.360 That's why I think, to begin with, it is important to include labels 00:52:24.440 --> 00:52:28.320 and for providers to do that, to start that education until people start 00:52:28.400 --> 00:52:32.680 to determine what is the right framework to evaluate. 00:52:32.680 --> 00:52:34.960 Because I think people are going to make things. 00:52:35.040 --> 00:52:38.080 And that's also why the provenance, the... 00:52:38.160 --> 00:52:39.640 What is it? 00:52:39.720 --> 00:52:42.520 Cppa, I'm going to mix it up, 00:52:42.600 --> 00:52:45.280 is so important because you're talking about derivatives, 00:52:45.280 --> 00:52:49.120 and that's where it gets really confusing, because if you create unique content 00:52:49.200 --> 00:52:51.720 that gets shared, somebody scrapes that content, 00:52:51.800 --> 00:52:54.760 they then create a derivative, and then someone else creates a derivative. 00:52:54.840 --> 00:52:57.640 It's like, how do you trace that lineage? 00:52:57.640 --> 00:52:59.400 And then who's responsible for that? 00:52:59.480 --> 00:53:00.520 And I think for And now, 00:53:00.600 --> 00:53:04.680 if you are a distributor, whether it's 1P or 3P, you include a label. 00:53:04.760 --> 00:53:10.960 And it might feel clunky at first, but at least we're doing that to support that. 00:53:11.960 --> 00:53:14.280 I understand. 00:53:17.440 --> 00:53:18.640 Hi, everyone. 00:53:18.640 --> 00:53:19.880 I just had a quick question. 00:53:19.960 --> 00:53:27.760 I think beyond labeling, what can consumer education look like with respect to AI? 00:53:27.840 --> 00:53:32.280 I feel like we're in a space A place culturally where we're 00:53:32.360 --> 00:53:38.480 not really focused on learning just what we're using with social media and whatnot. 00:53:38.560 --> 00:53:43.880 And as obviously everyone has background in terms of what AI is doing, 00:53:43.960 --> 00:53:51.080 I'm just wondering on the consumer side of things, how can we educate consumers on 00:53:51.160 --> 00:53:59.720 how companies are actually using AI with respect to their data? 00:53:59.920 --> 00:54:01.000 To that. 00:54:01.080 --> 00:54:04.360 Yeah, I think this goes also about... Sorry. 00:54:04.440 --> 00:54:07.920 My perspective on this is around, 00:54:08.000 --> 00:54:14.080 I mentioned the Nux, the new user experience, and how you meet them. 00:54:14.160 --> 00:54:18.440 You need to have engagement and education. 00:54:18.520 --> 00:54:23.440 If you meet somebody with a wall of text, it's like, Oh, this is scary. 00:54:23.520 --> 00:54:28.920 I think the tone of the conversation today is all about personalization. 00:54:29.000 --> 00:54:32.360 In order to achieve that, that means that the data... 00:54:32.440 --> 00:54:35.240 This AI needs to know you. 00:54:35.320 --> 00:54:37.160 I think that there needs to be 00:54:37.240 --> 00:54:44.840 an understanding of the value exchange and that data is encrypted and how it's 00:54:44.920 --> 00:54:52.040 used to train models and the sense making that's being made by the LLM versus 00:54:52.120 --> 00:54:57.280 it's not like somebody's defined how to identify you, 00:54:57.360 --> 00:55:00.680 but how do you demystify that with such small real estate? 00:55:00.680 --> 00:55:02.400 If you're also starting to talk about 00:55:02.480 --> 00:55:06.280 different interaction paradigms where it's not just with the screen and you don't 00:55:06.360 --> 00:55:11.040 have that affordance and maybe it's in a different environment. 00:55:11.120 --> 00:55:12.520 I'm not a gamer. 00:55:12.600 --> 00:55:14.200 Are there any gamers out there? 00:55:14.280 --> 00:55:16.240 Yeah, who Okay, I need to try to become a gamer. 00:55:16.280 --> 00:55:17.480 I feel like it's like... 00:55:17.560 --> 00:55:21.880 But there's a ton of generative AI experiences inside of games, 00:55:21.880 --> 00:55:24.240 but it's like, you're not going to be like, Oh, hold on. 00:55:24.320 --> 00:55:28.920 This couch that you're about to sit on, that was actually made by a machine. 00:55:29.000 --> 00:55:36.240 I think it's upon all of us to understand how these things are created. 00:55:36.320 --> 00:55:42.400 When you do have user experiences, how are you providing the affordances to educate? 00:55:45.960 --> 00:55:50.160 Hi, Bjorn Ostrad on my third AI revolution. 00:55:50.240 --> 00:55:53.520 I've been added for a while and found 00:55:53.600 --> 00:55:56.520 myself hilariously future-blind on all of them. 00:55:56.600 --> 00:56:01.080 I'm wondering what you think the coolest professions will be 10 years from now? 00:56:01.160 --> 00:56:05.040 Is that going to put the eye head for this question? 00:56:06.120 --> 00:56:10.920 Well, I mean, obviously, meta-employee into an employee runway. 00:56:11.440 --> 00:56:13.960 No, I mean, I think it's an excellent question. 00:56:14.040 --> 00:56:21.240 I think the reality is that the barrier to tools is just descending really sharply. 00:56:21.320 --> 00:56:23.680 I think Runway is a great example. 00:56:23.760 --> 00:56:26.400 We think about the ways in which 00:56:26.480 --> 00:56:30.920 access to information is just becoming easier easier and easier. 00:56:30.920 --> 00:56:33.280 And obviously, we have the standards and provenance pieces to balance. 00:56:33.360 --> 00:56:38.720 But your ability to have a personal tutor, 00:56:38.800 --> 00:56:45.760 then companion, then co-creator, co-ideator for anything that you want 00:56:45.840 --> 00:56:50.600 to try your hand at is really astonishing and really, truly new. 00:56:50.600 --> 00:56:53.480 And I think when it comes to AI revolutions, third time's a charm. 00:56:53.560 --> 00:56:58.080 So I think that the question is less about professions and more about tools. 00:56:58.160 --> 00:57:04.560 And I think that the tools are becoming more general and more imaginative and are 00:57:04.640 --> 00:57:08.440 more capable of augmenting human creativity and innovation in a way 00:57:08.520 --> 00:57:11.480 that makes your question excellent but impossible to answer? 00:57:11.560 --> 00:57:13.920 I have a short answer that I think being 00:57:14.000 --> 00:57:18.880 a philosopher, I think humans have ethics and we are very 00:57:18.960 --> 00:57:22.480 good at reasoning, whereas machines are good at reckoning. 00:57:22.560 --> 00:57:25.320 So I think the other question is, 00:57:25.400 --> 00:57:29.560 what's the future of academic institutions in 10 years? 00:57:29.640 --> 00:57:33.680 Is maybe the harder question to answer. 00:57:33.760 --> 00:57:38.680 I think professions are blending, 00:57:38.760 --> 00:57:44.120 so it's going to be hard to put them in a box in some way. 00:57:44.200 --> 00:57:45.440 We are seeing at a runway, 00:57:45.520 --> 00:57:49.520 we're seeing graphic designers turning animators, we're seeing animators turning 00:57:49.600 --> 00:57:54.600 filmmakers, we're seeing photographers turning writers, we're seeing this blend, 00:57:54.680 --> 00:57:58.120 specifically on the creative world, this blend of I am this, 00:57:58.200 --> 00:58:03.120 but I'm also this now because As we were talking about enhancing 00:58:04.200 --> 00:58:09.960 creativity or empowering people to do all the stuff, we're seeing I'm not... 00:58:10.040 --> 00:58:12.720 I come from a design bunker. I'm not a designer anymore. 00:58:12.720 --> 00:58:14.120 Maybe I'm also a game designer, 00:58:14.120 --> 00:58:15.720 and maybe I'm also a programmer or maybe or something else. 00:58:15.800 --> 00:58:23.120 So I'm not sure if I will qualify to one specific profession on its own. 00:58:23.200 --> 00:58:24.960 One thing that I'm super excited 00:58:25.040 --> 00:58:28.320 that I believe it will happen in terms of where we're talking about building 00:58:28.400 --> 00:58:33.440 context, and context in some way can refer to taste. 00:58:33.520 --> 00:58:40.240 And once you are building this things, 00:58:40.270 --> 00:58:43.520 specifically storytelling, you can start thinking on context being 00:58:43.600 --> 00:58:48.720 worlds, and worlds can be like, I want to create maybe a story or maybe I 00:58:48.770 --> 00:58:50.920 want to create a brand campaign for a brand. 00:58:51.000 --> 00:58:56.480 What is the context that I need to curate, accumulate, generate mix from different 00:58:56.560 --> 00:59:00.760 pieces in order to eventually create outputs of that world? 00:59:00.840 --> 00:59:05.360 That work can be a system, and that system allows me to create a website for a brand. 00:59:05.440 --> 00:59:08.920 That system allows me to create multiple 00:59:09.000 --> 00:59:12.880 video outputs or stories or podcasts out of that world. 00:59:12.960 --> 00:59:15.080 I call that world designers. 00:59:15.080 --> 00:59:18.600 I don't know if that's going to stick for long, but I am trying to coin that. 00:59:18.680 --> 00:59:23.800 And I think someone that will-Coin here at Neuhaus. 00:59:23.880 --> 00:59:27.640 Someone who will add this context to what 00:59:27.640 --> 00:59:29.200 they're going to be creating in the future. 00:59:29.280 --> 00:59:33.200 I'll take his wide one and go narrow. 00:59:33.280 --> 00:59:36.960 In my head, it's something like an investigative data journalist. 00:59:37.040 --> 00:59:43.720 I think that the more we use models, it's just going to compound over time. 00:59:43.720 --> 00:59:45.440 You're going to take this data and it's 00:59:45.440 --> 00:59:49.360 compounds, and then you could add a model or an app or a service on top of that. 00:59:49.440 --> 00:59:52.200 And all of a sudden it's like, where did we come from? 00:59:52.280 --> 00:59:54.440 And I think we're going to want to know that. 00:59:54.520 --> 00:59:59.720 And obviously, my background as a journalist, so that's how my brain thinks. 00:59:59.800 --> 01:00:02.720 But I do think that even... 01:00:02.800 --> 01:00:06.360 I was talking to somebody who's doing some work in the world of media, 01:00:06.440 --> 01:00:11.480 and I was like, how are you going to be doing fact-checking in this world? 01:00:11.560 --> 01:00:15.440 So the investigative element of who are we? 01:00:15.440 --> 01:00:17.560 Where do we come from? Where did you get that from? 01:00:17.640 --> 01:00:22.200 I think it's going to be a lot more important than we think. 01:00:22.280 --> 01:00:24.320 I'm actually hopeful in some of my former 01:00:24.400 --> 01:00:26.920 life, I was an entrepreneur, small business owner. 01:00:26.920 --> 01:00:29.040 Are there any small business owners in the house? 01:00:29.120 --> 01:00:30.400 Self-employed? 01:00:30.480 --> 01:00:32.080 Gig economy. 01:00:32.160 --> 01:00:36.720 Alejandra up here is an entrepreneur. I just... 01:00:36.720 --> 01:00:38.520 Everything that... Well designed. 01:00:38.600 --> 01:00:45.040 But I think if I can create a video story now, if I can run my marketing campaign I 01:00:45.120 --> 01:00:48.640 mean, dear God, if I don't have to do accounting and taxes, 01:00:48.720 --> 01:00:54.400 if all of that becomes taken care of and the means to those tools in production 01:00:54.480 --> 01:01:00.040 is really cheap, cost-effective, I just look forward to 01:01:00.120 --> 01:01:04.360 everyone starting a business or working for themselves. 01:01:04.440 --> 01:01:06.840 I think it'll be incredible. 01:01:07.480 --> 01:01:10.440 So I'm hopeful for that. Two more. 01:01:10.520 --> 01:01:13.600 We have time for two more? Two more. 01:01:13.680 --> 01:01:14.960 Two more. Okay, great. 01:01:15.040 --> 01:01:16.400 Who's next? Yeah. 01:01:16.480 --> 01:01:18.240 Someone behind a pillar that I can't see. 01:01:18.320 --> 01:01:20.040 Oh, sorry. Hi. 01:01:20.120 --> 01:01:27.160 So when we talk about Emotive AI and in context of traditional 01:01:27.240 --> 01:01:32.600 family norms and stuff, when it comes to kids or high schoolers 01:01:32.680 --> 01:01:38.360 that have this agent or a copilot or something that's guiding them through life 01:01:38.440 --> 01:01:43.560 instead of necessarily their parents, has there been any thought 01:01:43.640 --> 01:01:50.680 into the implications of how that could restructure just society in general? 01:01:50.760 --> 01:01:54.880 If you have a figure 01:01:54.960 --> 01:01:59.520 that is not necessarily your parental structure, but something that is guiding 01:01:59.600 --> 01:02:07.360 you that's overriding the advice that is being given to you by a parent? 01:02:07.840 --> 01:02:10.720 Yeah, I haven't really thought about it, but I'm sure it's going to be fine. 01:02:10.800 --> 01:02:12.320 Yeah. 01:02:12.400 --> 01:02:13.840 Next question. 01:02:13.920 --> 01:02:17.240 No, I think what you're What we're waiting on is really the... 01:02:17.320 --> 01:02:19.080 If we're at day zero, 01:02:19.160 --> 01:02:24.520 that's a very day one question to resolve, because right now we're getting to grips 01:02:24.520 --> 01:02:27.640 with and experimenting and pushing forward the capabilities of these models. 01:02:27.720 --> 01:02:29.600 But you're exactly right. 01:02:29.680 --> 01:02:32.640 I mean, you look at a good example is character. 01:02:32.720 --> 01:02:34.880 Ai, which you may be aware of, 01:02:34.960 --> 01:02:41.360 but it's a really simple idea where you are able to talk to these different agents 01:02:41.360 --> 01:02:44.120 that have been imbued with personalities by literally saying, 01:02:44.200 --> 01:02:48.480 You are Julius Caesar, teach me history, or You are Nikola Tesla, teach me science, 01:02:48.560 --> 01:02:54.920 or, You are my boyfriend now, and be my boyfriend, with that delightful voice. 01:02:55.000 --> 01:03:00.400 And folks are spending a ton of time, particularly younger folks, 01:03:00.400 --> 01:03:03.000 Gen Z and below, they're spending a ton of time engaging with these agents 01:03:03.080 --> 01:03:07.320 and forming relationships really free form and really Wild Westley. 01:03:07.400 --> 01:03:12.680 And I think that a discussion about where 01:03:12.760 --> 01:03:17.280 AI occupies roles and maybe shouldn't occupy roles in your life 01:03:17.360 --> 01:03:20.160 is a really critical, maybe the most critical subset 01:03:20.240 --> 01:03:23.040 of the conversation we're finally beginning to have about the role 01:03:23.120 --> 01:03:27.160 of technology in children's lives and young people's lives. 01:03:27.240 --> 01:03:28.680 Because we're really late to the party. 01:03:28.760 --> 01:03:30.120 It's happening now. 01:03:30.200 --> 01:03:32.240 It's mostly through the lens of screen 01:03:32.320 --> 01:03:37.520 time and content on whatever platform that you're consuming content on. 01:03:37.600 --> 01:03:41.960 But yes, as you rightly point out, as we begin to reach a point where we can 01:03:42.040 --> 01:03:45.800 develop relationships, invest emotion into, and ultimately, 01:03:45.880 --> 01:03:51.320 psychologically depend upon agents in AI systems to some degree, yeah, 01:03:51.320 --> 01:03:52.960 that's going to need to be a really active conversation. 01:03:52.960 --> 01:03:54.720 But one, I don't think we're even... 01:03:54.800 --> 01:03:58.120 I think it's day one and we're at day zero still. 01:03:58.200 --> 01:04:04.200 I have a an answer to that, which is, first of all, I love the question. 01:04:04.280 --> 01:04:10.560 I think that I'm going to pick on to Carly's philosopher point here, 01:04:10.640 --> 01:04:16.200 which is that if, let's say, we have to have parental control for these 01:04:16.280 --> 01:04:19.600 agents, much like we was talking about screen time. 01:04:19.680 --> 01:04:23.000 I think what it's going to make parents do 01:04:23.080 --> 01:04:28.600 is think about the value system that they need to imbue those agents with. 01:04:28.680 --> 01:04:33.280 And so then the question is going to be to every parent, what is your value system? 01:04:33.360 --> 01:04:37.200 What are the actual controls that you're going to put onto that agent? 01:04:37.280 --> 01:04:41.800 I don't know the parents think about that today, but I think tomorrow they're 01:04:41.880 --> 01:04:46.080 going to have to think about the values they want their kids to grow up with. 01:04:46.160 --> 01:04:49.120 I guess we have time for one more. 01:04:49.200 --> 01:04:51.120 One more question. 01:04:52.800 --> 01:04:53.760 Thank you. 01:04:53.840 --> 01:04:59.400 I'm a software engineer, and I'm really interested in the therapy industry. 01:04:59.480 --> 01:05:03.720 Actually, In my leisure time, I provide 01:05:03.800 --> 01:05:07.120 consulting, which is like the life coaching for my friends. 01:05:07.200 --> 01:05:12.840 And my question is, how will the emotive AI change the therapy industry? 01:05:12.920 --> 01:05:14.200 Will it replace 01:05:14.280 --> 01:05:21.800 the therapist and the psychologist, or at least take some share from that market? 01:05:22.280 --> 01:05:24.920 Yeah, that's an excellent question. 01:05:25.000 --> 01:05:27.240 I think bottom line, 01:05:27.240 --> 01:05:29.640 and one of the ways that I think it's really important to think about AI 01:05:29.720 --> 01:05:34.560 and of the mode of AI is that we still need to maintain agency in the person. 01:05:34.560 --> 01:05:35.880 So when I'm thinking about therapeutic 01:05:35.960 --> 01:05:40.200 applications, I think mood journaling and treating the AI as 01:05:40.280 --> 01:05:44.960 a way to express and extend what you're feeling into a way that's 01:05:45.040 --> 01:05:49.920 you can reflect upon, your therapist, your trained professional can reflect upon. 01:05:50.000 --> 01:05:52.520 That's certainly where we think about it, 01:05:52.600 --> 01:05:56.560 hope that it's heading, and what we're pushing for is that making 01:05:56.640 --> 01:05:59.840 sure we're not creating negative dependencies, but instead, 01:05:59.920 --> 01:06:05.320 encouraging people to maintain, but not just maintain, increase their agency. 01:06:05.400 --> 01:06:09.040 So that's self-expression, self-reflection. 01:06:09.120 --> 01:06:13.880 That's going to be super important to, again, obviously standardize and regulate. 01:06:13.960 --> 01:06:16.640 But we hope that it takes humans 01:06:16.720 --> 01:06:20.360 and doesn't put them in a less empowered role, but actually a more empowered role 01:06:20.440 --> 01:06:25.320 to, again, self-express, self-reflect, and hopefully heal in the case of therapy. 01:06:25.400 --> 01:06:28.880 I think it goes back to the centar system, though, and human in the loop. 01:06:28.880 --> 01:06:32.880 Like, lonely This was a big problem during the pandemic, access to mental health. 01:06:32.960 --> 01:06:38.680 I think a big part of self-care is social connections. 01:06:38.760 --> 01:06:45.160 I think that therapists in person provide a different dimension than remotely. 01:06:45.240 --> 01:06:47.240 I haven't looked at the data, 01:06:47.320 --> 01:06:50.720 but I do think it just goes back to that idea of, I think you even said 01:06:50.720 --> 01:06:53.520 human in the loop, and I think a call to center system is so 01:06:53.600 --> 01:06:58.640 much more important than just an emotive therapist on its own. 01:06:58.720 --> 01:07:04.000 I think I'll make just a call back to the question about cool jobs of the future. 01:07:04.080 --> 01:07:07.000 This is technology that is really going 01:07:07.080 --> 01:07:11.360 to demand a deeper level of understanding about humans. 01:07:11.440 --> 01:07:14.600 You're not going to A/B test your way to the best design. 01:07:14.680 --> 01:07:16.920 It's not going to happen. This is not a thing you're just going 01:07:17.000 --> 01:07:21.520 to pop up on user interviews and get some feedback and know what to build. 01:07:21.600 --> 01:07:26.800 Some real deep anthropological, sociological, psychological understanding 01:07:26.880 --> 01:07:30.560 of what people need when and how to treat ethically. 01:07:30.640 --> 01:07:33.560 I would piggyback on philosopher and add ethicist. 01:07:33.640 --> 01:07:36.120 We need more data ethicists in the world. 01:07:36.200 --> 01:07:38.720 But I think this is a place where really 01:07:38.720 --> 01:07:42.680 getting to know people is going to be more important than ever because we're trying 01:07:42.760 --> 01:07:48.480 to substitute an artificial intelligence for human interaction in certain moments. 01:07:48.560 --> 01:07:52.280 If it's going to work, it's going to need to feel really authentic. 01:07:52.360 --> 01:07:55.840 With that comes all the scary potentialities. 01:07:55.920 --> 01:08:01.680 But in any event, all really good good questions and scary 01:08:01.760 --> 01:08:06.000 questions, some of them, but also really optimistic potentials for all of us here. 01:08:06.080 --> 01:08:08.200 I want to thank everyone. 01:08:08.280 --> 01:08:12.320 I want to thank our panelists for an excellent conversation. 01:08:12.400 --> 01:08:14.840 There are drinks to be had and more 01:08:14.880 --> 01:08:19.080 conversations to be had and this wonderful space to take advantage of out of the rain 01:08:19.160 --> 01:08:22.280 and under a giant disco ball or something over there. 01:08:22.360 --> 01:08:24.160 So thank you all. 01:08:24.160 --> 01:08:25.240 Thank you to the audience. 01:08:25.240 --> 01:08:26.440 Thank you to our panelists. 01:08:26.520 --> 01:08:30.200 And thank you to us, too, at Noya House for hosting us. 01:08:30.280 --> 01:08:33.040 Yeah. Thank you.