TrueValueMetrics ... Impact Accounting for the 21st Century

Date: 2025-08-21 Page is: DBtxt003.php txt00026833

EVENT TRANSCRIPT ... JUNE 5th 2024 Axios AI+ NY ... in partnership with Tech:NYC Axios is bringing our AI+ Summit to NYC for Tech Week Hear from leaders who are shaping the future of AI, from finance to media and healthcare & more.
992 views Streamed live on Jun 5, 2024 Axios is bringing our AI+ Summit to NYC for Tech Week in partnership with Tech:NYC. Hear from leaders who are shaping the future of AI, from finance to media and healthcare & more. Music 5 songs In Corporate Enrize Enrize All to Myself (Instrumental Version) spring gang All to Myself A Minute Too Early Arc De Soleil Lonely Party The Main Event Matt Large The Main Event Calling All Discos Andreas Dahlbäck Calling All Discos / Maybe We Meet in November Music Transcript\ 0:09 [Music] 0:15 [Music] 0:21 [Music] 0:35 [Music] 0:57 [Music] 1:15 [Music] 1:21 [Music] 1:27 [Music] 1:41 [Music] 2:03 [Music] 2:27 [Music] 2:33 [Music] 2:47 [Music] 3:08 [Music] 3:27 [Music] 3:33 [Music] 3:39 [Music] 3:52 [Music] 4:14 [Music] 4:39 [Music] 4:45 [Music] 4:58 [Music] 5:20 [Music] 5:39 [Music] 5:45 [Music] 5:51 [Music] 6:04 [Music] 6:26 [Music] 6:45 [Music] 6:51 [Music] 6:58 [Music] 7:10 [Music] 7:32 [Music] 7:51 [Music] 7:57 [Music] 8:03 [Music] 8:16 [Music] 8:38 [Music] 8:56 [Music] 9:03 [Music] 9:09 [Music] 9:22 [Music] 9:44 [Music] 10:02 [Music] 10:09 [Music] 10:15 [Music] 10:28 [Music] 10:50 [Music] 11:08 [Music] 11:15 [Music] 11:21 [Music] 11:34 [Music] 11:56 [Music] 12:14 [Music] 12:21 [Music] 12:40 [Music] 13:02 [Music] 13:20 [Music] 13:27 [Music] 13:46 [Music] 14:08 [Music] 14:26 [Music] 14:32 [Music] 14:38 [Music] 14:52 [Music] 15:14 [Music] 15:32 [Music] 15:38 [Music] 15:44 [Music] 15:58 [Music] 16:19 [Music] 16:38 [Music] 16:44 [Music] 16:50 [Music] 17:03 [Music] 17:25 [Music] 17:44 [Music] 17:50 [Music] 17:56 [Music] 18:09 [Music] 18:31 [Music] 19:02 [Music] 19:15 [Music] 19:37 [Music] 20:02 [Music] 20:08 [Music] 20:21 [Music] 20:43 [Music] 21:02 [Music] 21:09 [Music] 21:43 [Music] 22:13 22:38 welcome aboard [Music] to kick things off please welcome Axios Chief technology corespondent Ena Fried hi every everyone. Wwelcome it's so great to see you all I'm both here and on the live stream I'm excited this is our third AI plus Summit and our first in New York City and what better time to have it than during New York's Tech week and first off I really want to give a big shout out for the contributions of Tech NYC both for helping us with this and for all the events they're doing during Tech week we launched the AR plus Summits in November 2023 we really noticed as I think everyone did that Ai and particularly this generative AI wave was going to be the most consequential shift in a long time it's certainly the most consequential shift in my 20 plus years of covering Tech and we wanted to have a regular venue to get into the details to talk about what's working what's not and I really think of today as a check-in and we really should look hard at what is working ... what we want to see more of ... what we we need to worry about because there's a ton of fears around AI some of which we probably should be worrying about even more some of which we might need to worry about in a little bit and some of which probably aren't the right areas to place our concert so we're going to talk about all of that today um we also have a few surprises in store I'm not going to spoil too much but there will be a robot and some football gear that's all I'm going to say about that um to give you a quick road map of today the mainstage program will run until around 5 20 when we'll have a cocktail reception right here immediately following so don't forget um to stick around for that you can post all about today on social use the hash axios AI Summit um feel free to venture downstairs if you need uh to use the restroom take a phone call whatever Refreshments um but all the programming will be happening up here and I want to mention one fun thing we're doing which is with our partners at meta uh where they're showcasing their AI image generator imagine and one of the things that different about imagine is um you 25:01 know a lot of these text to image engines you type in something you get back an image you don't like it you type in something new what imagine lets you do is it actually starts generating as you're typing um and that's both saves you a little time but what I really like is it kind of gives you Insight I'm hesitant to call it how it's thinking because I think we do have to be careful about anthropomorphizing tech but you really do see how it works so we want all of you if you're interested to to give it a try um there's instructions on the back of your um cards your name tags um with a QR code scan that um and as far as topics um maybe something along the lines of what might New York City look in 20 years or what's a perfect New York day um if you enter if you submit 25:48 something um you'll be entered to win one of a couple pairs of the Rayband 25:54 meta smart glasses which I've been using both to record my softball games I also interviewed Sam Altman while live 26:00 streaming from them um so those are really cool um lastly I want to remind folks that we are going to be doing our West Coast installment of the AI plus Summit December 17th in San Francisco which happens to be my 50th birthday so we're going to party for a number of reasons we'll have survived this year which will be something to celebrate um without further Ado I want to welcome two of aio's co-founders the axio CEO oh sorry um axos uh Mike Allen and axios HQ co-founder and president Roy 26:35 Schwarz thank you for the promotion and for your coverage happy birthday a little 26:42 early uh hello to axio AI plus Summit uh both in the room and virtually around 26:49 the world I'm Mike Allen a co-founder of axos and we're going to do something we've never done before for the summit I 26:57 and my co- founder Roy Schwarz who's also the CEO of aus HQ our AI platform 27:04 for communications are going to go inside the conversations that we've had every morning with Jim vahi for going on 17 years now 17 years 4:30 a.m. feel free to join us now Roy were you a axos early bird or a natural early bird I'm definitely not a natural early bird but these guys wake up super early so they figur they'd give me a call and wake me 27:27 up uh so uh we're going to unpack three quick things here to to kick things off 27:34 at the summit how we're thinking about talking about AI in the world how we're thinking about it in media and how we're 27:41 thinking about it in Communications to start with the world so we were rehearsing a little bit ago and exus 27:47 publisher Nick Johnston who you'll hear from in a bit said everyone's here 27:52 because we're fascinated and we're terrified and I was like that is perfect yeah exactly and it's also we want to 27:59 know what's real right what's going to lead to True productivity there are billions of dollars being spent how is 28:04 that going to translate into what we do in our work lives what we do in our personal lives yeah and try to get past what's real what's hype uh what is uh in the wild and most importantly a sneak peek at what's coming when we think about axios in media and this is something that Jim UI have been obsessed with for years now is how is gener AI 28:26 going to affect what we do and how can we use it to bring even more expertise 28:31 to our audience to make smart professionals smarter faster on the topics that matter most especially 28:37 including Ai and the way that we look at it is one like we're leaning into being 28:43 Innovative and experimenting but two in our Newsroom it's always humans first 28:50 journalists first which is part of how we build trust with our audience and so 28:55 and we're using the tool uh to help our journalists do more on 29:01 more topics for more people yeah one of my favorite use cases for how we're using it in media is we take data sets 29:07 where we have you know all this city data we have 30 different cities that we Now cover on the ground locally and we 29:14 grab that data set and we're able to quickly generate some drafts of content using that data in order to help the 29:21 journalist get a head start start well we very much take a trusta verify uh approach and something that Jim vandah 29:27 High RC has said is that in these times there will be even more of a premium on people 29:33 like Ena Fred who can tell you things that you don't know like that's always going to be uh the gold standard now Roy 29:40 after getting uh axios going so uh Roy Schwarz who became a close friend and a 29:47 collaborator at Politico uh before long before actually had a name we were uh Jim you and I 29:55 looked at a whiteboard uh in the in the break room at part bu building and dreamed up what would become axos and 30:04 you're now over at axos HQ where you're spending all your days talking to some of the most interesting people in the world corporate leaders experts about generative Ai and Roy what are you hearing in those conversations about how people differen makers are actually using AI yeah there's there's two things that I've now seen in talking to Executives in terms of how they're actually used utilizing AI right now uh one of those is as a knowledge base so searching through emails documents 30:35 spreadsheets being able to extract information to understand where can I 30:40 find this answer to this question amongst all the corporate documents so that's real that's definitely happening 30:45 right now the other is in augmenting writing so I see a lot of tools that are helping people write more that 30:52 unfortunately means a lot more emails a lot more text messages coming your way in the next few years uh but that's also 30:58 our focus at axos HQ how do we make everyone better writers better communicators and make everything sort 31:04 of more efficient at work now Roy in the spirit of going behind the curtain and a sneak peek for this audience here in the meat packing District of New York and on the live stream around the world what can you tell us about tools that are coming that will help uh people with these big problems yeah we have something to show you today it's a world premiere no one's seen this before so this is is our version of co-pilot we call it comms pilot um and we're going 31:32 to play a short one minute video comms pilot is the new virtual 31:40 Communications assistant coming to axio HQ it combines the editorial Excellence of axios and the communications 31:46 expertise of HQ to improve internal comms let's take a quick look at what comm's pilot can do comms pilot can 31:54 quickly craft a smart bity card on any topic important to your organiz ganization using your draft copy or even 32:00 just suggestions it can analyze long content and turn it into something quick and easy to read it can then work with 32:07 you to fine-tune that update it can also add more or edit the details down to 32:12 what your readers care about most and over time it learns more about your organization every time you use it it 32:19 can look at and analyze your past Communications and remind you about things you haven't touched in a while or 32:25 a project it's time to share updates on that's really cool I saw that Jimmy who 32:31 was the voice that gave himself a shout out in there and I left the little brainstorm uh button it's very cool uh 32:38 very friendly so uh to finish up here we always at axius we always look at what's 32:43 next and looking ahead three quick ideas that Jim you and I think about how can 32:50 AIO serve this amazing audience better in an AI world the first is trust and 32:56 that's always going to be the number one thing for us that uh clinical coverage 33:01 no bias no opinion never waste your time never insult your intelligence and the 33:07 second this is Ena freed the expertise human expertise which has been part of 33:13 the design from axios from day one but more important than ever yeah and now at 33:18 ax HQ when we think about humans in the loop we want to make sure that it's not just the AI we're also using best 33:24 practices and expertise so we are actually using former journalist uh in fact one of our people is the 33:29 former writer of the presidential Daily Briefing in terms of how do you write content how do you frame content and we 33:36 use that expertise in order to help the recommendations for the AI and the third part of how axios is thinking about the 33:43 future in an AI world is something that you all are a part of both here uh on 33:49 set and uh people watching around the world and that is human connection which 33:55 has never been more important an expert uh talking to one of our 34:00 journalists vice versa telling you something you don't know Illuminating narrating uh the world so uh very 34:07 excited about the conversations ahead and uh Roy what do you have for us uh we 34:13 do have a demo in the back so if anyone wants to see AI live can go back anytime uh we'll be happy to show it to you and 34:19 I think on with the show uh great uh we thank IBM and meta for making these conversations possible the incredible 34:26 axio events team which is here all night spinning up this amazing uh experience 34:32 thank all of you for being the greatest audience in the world have a great AI plus Sumit thank 34:38 [Applause] [Music] 34:47 you welcoming back Ena [Music] 34:53 freed all righty to kick things off our first guest is is a researcher whose 34:59 background is steeped in National Security knowledge and major policy decisions and how AI is affecting all of 35:05 that until recently she was a director and board member of open Ai and in her recent Ted Talk which I caught in 35:12 Vancouver she called for greater transparency and auditing of how AI is being used by big Tech please join me in 35:19 welcoming Helen toner [Applause] [Music] 35:28 on the Ted aai podcast that dropped last week you shared a little bit more about the board's rationale for firing Sam 35:35 Alman you mentioned a couple things not getting a heads up before chat GPT launched finding out like most of us did 35:42 on Twitter not knowing about some of his ownership involvements with a fund um 35:47 that invests in related companies um and I I think also you you felt that you were being unfairly targeted for a paper 35:54 You' written is that the main things or what what else was kind of you know that you 36:00 can talk about was going on in that decision yeah unfortunately there's still a lot that that is pretty hard to 36:05 talk about just for all kinds of legal and confidentiality reasons but um the way I would put it very briefly is 36:10 essentially that we had had years worth of issues with trust accountability oversight and then that was really 36:16 compounded by some very serious conversations we had with leadership at the company um a couple of senior 36:22 Executives last fall um some of the examples you named were part of it there are many other examples that formed a 36:28 larger part of the picture um I think better not to kind of rehash it all here but if you know folks are interested 36:34 they can certainly go look at that podcast episode um and uh I think you know maybe we can leave it there for for 36:39 now the rationale that you shared at the time that the board shared was that Sam 36:45 wasn't being consistently candid um and it was interesting because at the time I think you know everyone was trying to 36:50 read the tea leaves at the time open AI was flying high there was really no indications of any sort of turmoil from 36:57 the outside and it was hard to understand that idea of not being consistently candid when you flash 37:03 forward to now we've had Scarlet Johansson in the talks and then they went ahead with a voice that sounded 37:09 like her even though there was no deal we had um former employees saying well 37:14 we can't really talk because we had to sign these non-disparagement Clauses Sam saying oh I didn't even know about them 37:20 but then if you look at the letters Sam was the one who' signed those it seems like the board's arguments might have 37:27 landed Ed a bit differently if they were happening now after all this do you wish 37:33 you'd waited a bit I think there's a lot of reasons we did it the way that we did and I don't think there are big changes 37:39 that I would make going back um just because of a lot of a lot of different factors that going on at the time 37:45 unfortunately still you know really hard to kind of get into all the details of them um I think what we've seen over the 37:50 last few months though is really increasing awareness and um I think you know public understanding of the 37:56 importance of governance for these companies the importance of the processes they're using to make decisions uh the importance of trust in 38:03 the companies and the leadership and I think that's only going to you know intensify as as AI keeps getting more 38:08 sophisticated now you you and other folks said you know it wasn't you didn't see some you know robots taking over 38:15 behind the scenes it wasn't some giant safety fear that led you to take what you want at the same time it does seem 38:22 like there are a lot of concerns around safety we've heard a lot from uh whether it was Yan when Ilan Yan left we've 38:29 heard more in recent days there's an open letter saying safety's not being taken seriously enough and not just 38:36 about open AI specifically but in general what should we be talking about when it comes to open sorry when it 38:42 comes to AI safety yeah I think you know there's a ton of different issues AI is a very very broad space lots of 38:49 different kinds of safety governance ethics issues we could talk about I think um for a small subset of companies 38:55 there's open AI but also anop OPC Google meta that are explicitly trying to build extremely Advanced AI systems some 39:02 people will call it AGI artificial general intelligence um other people don't like that term myself kind of included um but I do think for that for 39:10 those companies that are explicitly trying to build very very Advanced AI systems that is a really ambitious goal 39:16 it is a goal they will all tell you you know stands to potentially Change the World um and a goal that many 39:21 researchers tell will tell you comes along with really significant risks so again I think something that we're 39:26 starting to see a little bit more awareness of over the last few months and years is the fact that you know if 39:32 you're a company and you're trying to do that that's your stated goal I think you should be subject to a little bit more scrutiny um a little bit more um you 39:39 know questioning and and uh the public the government wanting to understand kind of what is it that you're doing and 39:44 and why do you think that it's not just safe but also you know responsible going to you know redown to the benefit of of 39:50 society and the public and not just to the benefit of you know specific individuals at those companies um so I 39:56 think there's yeah again kind of almost too many different AI safety issues to try and Tackle them all but I do think 40:01 um there's a particular set that apply to those companies that are explicitly trying to build very Advan systems and I just want to highlight how unusual the 40:09 set of circumstances are that we're talking about here because I think what you're calling for is rather 40:14 unprecedented so in your Ted Talk You' called for greater auditability but also 40:19 that maybe these companies should have to share what it is they're working on now tech companies historically have not 40:25 been sharing what they're up to too but you know and we've talked about this this is an unusual moment where the 40:33 absolute Leading Edge of the technology is not being developed by governments it's being developed by private 40:40 companies what what is reasonable to ask of private companies I mean they're not 40:46 going to want to share everything they're doing but what they're doing could affect all of humanity how do we 40:52 how do we deal with that and who's in charge cuz putting governments in charge is not always the best either absolutely 40:58 not I mean I think the kinds of things that the kinds of policy interventions that I think make sense right now are actually not that unprecedented not that 41:05 drastic um and I think they really come from my big picture of you here is that 41:10 with AI we're in a place of huge uncertainty so if you talk to different experts about how they expect the field 41:16 to develop over the next five or 10 years if you look at um you know the extent to which we've been surprised 41:22 over the last five or 10 years I think it's only reasonable to think that we shouldn't be confident one way or 41:27 another what AI will look like you know over the coming years and so you know for me that leads to being interested in 41:34 policy measures that can help us understand the field better um and that I think are pretty light touch honestly 41:39 I mean I think things like if you are spending $100 million training a Frontier Model you should have to run 41:45 some safety tests and you should maybe have to report um report those tests or the results of those tests that you know 41:50 that's that's really not all doing that they're all saying they're doing that they're saying that they're doing the 41:56 tests um but there's not a lot of standardization in what they're testing for how they're reporting the results um are they reporting the results at all um 42:02 I think there's been good progress on this but I think I think we need more um I think also things like having you know 42:08 third party independent Auditors being able to um verify some of the claims that the companies are making so that 42:13 you don't just have to trust kind of the word of the corporations themselves I mean these are not crazy regulatory 42:18 interventions we see much heavier things in you know if you look at something like food safety or you look at Auto 42:24 Safety like those are fields that don't stand to threaten Humanity and and it's very very normal that they're subject to regulation to make sure that uh that the 42:31 the products work as they're supposed to work and that we kind of understand what's going on so I think there's a little bit of a I think our expectations 42:37 around policy are a little bit wonky when it comes to Tech because we have had a few Decades of the tech sector 42:43 specifically being so lightly regulated relative to basically every other sector um I mean don't get me wrong there's 42:49 obviously lots of ways to regulate in a sort of clumsy or heavy-handed way that could could backfire and would not be 42:55 worth it um but I also think that uh sometimes when I talk to folks who are in the AI sector you know doing research 43:01 or building AI products they have this sense of any kind of Regulation is going to be the government coming in and 43:06 telling me what I can and can't do and crushing Innovation and handing our lead to China and you know it's all going to be terrible and I think there's just 43:13 plenty of places where that is not how regulation works and if you can um have smart people in government which I think 43:19 in this case we actually do have some pretty smart people working on AI in the US government um if you can be cautious if you can be iterative start small um I 43:26 think there's really good prospects to introduce some some really good good policy measures here and one of the things that's always been interesting to 43:33 me and I've always pushed for is you know what is our role as Citizens and you've talked some about that what can 43:40 people who care about this who are interested what is your recommendation for what what individuals can do to be 43:47 smart citizens to call for the right things to put pressure on companies to 43:52 behave differently what can individuals do yeah I think you know you mentioned the talk I gave um a month or two ago 43:59 and my kind of biggest suggestion there was to not be intimidated by this as a technology I think people hear Ai and 44:05 they think oh my goodness that's way too complicated I could never understand anything about it um and in some ways 44:10 you know there's some complicated math involved um but in other ways it's it's pretty straightforward um it's really 44:16 just computers trying to make predictions Based on data and um it's also such a wide space with so many 44:22 different implications and applications that no company no government a can kind of understand all the ways that 44:29 it's affecting our lives and so I sort of see there is almost like in the same way that you know a big advantage of 44:34 markets is they can handle lots of different realities in lots of different distributed places you don't have one 44:39 you know uh uh hegemonic um you know Soviet uh control in the middle um in 44:46 the same way with AI I think there's a role for citizens to be looking at how is this affecting my life how do I want it to affect my life um what would you 44:53 know what would I hope for my computer to be able to help me with versus how is it actually harming me or taking away 44:58 things that I enjoy um and just feeling entitled to that perspective and entitled to to your own experience I 45:04 think is sort of the biggest the biggest thing and then in terms of tactically you know there's all the regular things you can talk to your legislator you can 45:10 let companies know when they're doing things you don't want you can vote with your wallet etc etc um but I think the 45:16 the sort of core underlying thing is not to think of this as a distant mysterious magical technology that you could never 45:22 possibly kind of have a view on but instead to just you know have your own thoughts on how it's affecting you and your life and the lives of people that 45:29 you know and you know you given the actions that you've taken I think a lot of people perceive you as you know 45:35 against this sort of Technology you're not from what you've told me what do you 45:40 think the technology today is good for and what do you imagine it will soon be good for yeah I mean the what it will 45:48 soon be good for is kind of or guess depends what we mean by soon um I think in principle you know AI is uh a lot of 45:55 people think it could accelerate Science and Technology research more and progress more generally I absolutely 46:01 think that technological progress is one of the best things that's happened to humanity over the last few hundred years 46:07 few thousand years um so that Prospect I think is just very very broadly very exciting in terms of what it could mean 46:12 for our quality of life um in terms of kind of more near-term I mean personally 46:17 I think looking at what are big problems and how could AI help with those problems um is a little bit of a cliche 46:23 but you can make it very concrete and for me there's things as simple as you know autonomous driving where right now 46:29 something like 30,000 people die every year from cars like that's crazy that's 46:34 a huge problem that's just in the US um so if we can have and we already see in in San Francisco we have there's a 46:40 couple of companies where you can already hail a driverless taxi and so yeah I think sometimes people hear me 46:45 talking about like well policy and risk and whatever and they think oh you must be against all this stuff but like I'm so excited about autonomous vehicles 46:51 obviously we should test them carefully we should check their you know uh as safe as human drivers but if they are 46:57 that's such a concrete tangible really serious example of a problem that I think AI could could solve over the next 47:03 five or 10 years um so that's an example that I'm particularly excited about and in the immediate term do you use a chatbot do you use today's generative AI 47:11 I use it some um I think it can be really helpful for when you're drafting something and you want some different ideas or you want critique of you know 47:18 uh what could be clearer um personally my favorite use for it is when there's something on the tip of your tongue and 47:23 you can't quite get it um I find that chat Bots are really good you can like well it kind of means this and it kind of means that and I think it starts with 47:29 p but maybe it starts with t um they'll really help you find it quickly um also translation I do you know a chunk of my 47:36 work relates to kind of China's AI ecosystem and um I think just the the way that machine translation has 47:42 improved over the last 10 years has been absolutely astounding um and I'm I'm glad that I speak enough Chinese to go in and look at original Source documents 47:49 um but often it's very helpful to be able to skim in English and have a high quality you know English translation to be starting from and as you mentioned 47:55 there's there's a lot of of risks when it comes to a AI generative AI in particular um there's the existential 48:02 risks of you know literally the systems taking over but there's so many other 48:07 things whether it's amplifying existing biases misinformation wealth inequality 48:14 all these things how do we have a nuanced conversation I've always been a 48:19 big advocate of we got to put existential risk we got to have that conversation have it off to the side and 48:26 then we have to a conversation about other things that are not theoretical they're here now 48:31 misinformation all the ones that I talked about but we've actually debated this some and you you feel like we don't 48:37 have to separate those conversations why not and how should we have that conversation yeah I I think it's really 48:43 unfortunate that they've become like such separate conversations and such separate perspectives um because I think 48:49 a lot of the you know if you're talking about on the one hand you know a company using an AI tool for hiring and it's a 48:54 discriminatory tool and on the other hand you have uh you know an artificial general intelligence that recursively 49:00 self-improved to be a super intelligence like that can seem like very very different things I get it um but I also 49:07 think that if you're looking more broadly at uh I think there's a much many through lines between issues we're 49:12 seeing here and now and issues that we might need to anticipate different ways the technology could cause harm could go 49:17 off the rails um I think there's a lot of common threads so things like um how do we control and understand AI as a 49:24 technology or deep learning specifically which is kind of the type of AI that is you know most popular developing most rapidly right now um as you obviously 49:31 know uh you know the fact that we can't control and understand deep Learning Systems the same way that we you know 49:38 understand almost any other engineered system we have is at the root of a lot of these problems in terms of you know 49:43 what gets called the black boox problem we don't really know how they work we don't really know how they'll fail um also questions around kind of power 49:49 Market concentration who is making decisions who is benefiting I think are really common to not just questions of 49:56 algorithm bias or disinformation today but also could have really big implications over the next um next few 50:02 decades and could lead to sort of civilizational scale bad outcomes um I also think that sometimes people hear 50:07 the phrase existential risk and they just think kind of Skynet and robots with guns like shooting down humans and 50:14 I think there's a lot of other ways that um that AI could really take things in a 50:19 bad Direction humans could take or humans could use AI to take things in a bad Direction at a global scale at a 50:25 civilizational scale and so I think sometimes we um do ourselves a disservice by kind of locking in these 50:31 very specific stories of what we're talking about as opposed to thinking sort of more holistically as this is a very powerful technology we don't really 50:38 get how it works yet maybe we can you know later um and right now it's being developed and pushed forward by a very 50:44 small group of people and I think there's lots of different reasons to be concerned about harms that causes right now and different harms that could cause 50:50 as the technology progresses so we only have a little bit of time left and I kind of want to return to where we started in a sense you know one of the 50:57 unique things about opening I was they had this nonprofit be dominant in fact part of the mission was if things are 51:03 going in a bad Direction pull the rip cord my feeling was always the board only had one rip cord um and my sense 51:11 even after what happened is you guys kind of pulled that rip cord and now open AI has a very different structure 51:17 and a very different board what do you hope we all take away from that moment cuz I do feel like that was the rip cord 51:24 I don't have I don't know if you have confidence that the new board would act similarly if it saw things of concern 51:30 but I feel like the old open AI is gone do you agree and what should we have learned I definitely want to give the 51:36 new board a chance I think they have a lot of better structures in place that were not possible for us to um to put in 51:42 place uh before November um but I do think that more broadly it's a good demonstration of the power of profit 51:50 incentives the power of um just the the stakes of this technology and the way 51:55 that I think relying on corporations to self-govern and to make the right decisions even when it goes against the 52:01 interest of investors um or the interest of other kind of powerful parties is is not going to be enough and so again I 52:07 think to the extent that we think AI is going to be um a powerful technology that is going to really have big effects 52:13 for the world then we shouldn't just trust you know private actors to to get to decide what to do with it well I 52:19 think this is a great way to start what I hope it's going to be a fascinating day I know we're going to dive a bunch more into a lot of these topics um and I 52:26 appreciate all of your Insight thanks thank you very much Helen toner 52:34 [Music] 52:56 oh [Music] 53:32 welcoming our view from the top moderator Nicholas Johnston good afternoon everyone thank 53:40 you so much for being here uh I love our events a live representation of axio 53:46 Journalism this is our third time doing an AI event we couldn't do it without all of you who are here uh as our 53:52 readers and we thank you so much for being here we're also hugely hugely grateful uh to our partners who make 53:58 everything uh like this possible so a huge huge thanks to IBM for supporting today's event uh and before I jump in on 54:04 that conversation uh you're probably familiar with a little bit of what IBM does but let's do a quick video uh to go deeper on it I'll be right 54:11 back one more game Chris I don't know Eli I have to do some research for the US Open broadcast oh you should use a US 54:18 Open app for that it's full of AI generated insight to get fans and broadcasters up to speed on every match 54:23 like the IBM power index likelihood to win predictions and draw analysis it even uses AI to generate spoken 54:30 commentary for hundreds of match highlights okay then game on let's go AI generated commentary and insights built 54:37 with IBM Watson X so we've probably all seen some of 54:43 that stuff uh during the US Open during Masters and so I'm very excited to go a little bit deeper on it with two special 54:49 guests uh entrepreneur and Tennis Legend Maria Sharapova and The Man Behind the magic uh IBM's SVP for communic Jonathan 54:57 adek Maria jonan welcome to 55:02 axio uh so jonan I want to start with you not an empty seat in the room nice 55:07 not bad right like it yeah uh All For You Jonathan no no no no for you come on 55:13 uh it makes sense why IBM is here but you showed up with a tennis Legend friend uh and we've seen a lot on 55:19 television a lot what IBM does at the Masters US Open pretty cool stuff to do give us a view from your Vantage Point 55:25 about how IBM thinks about that and what you're focused on yeah well first thanks for having us and Maria thanks for for 55:31 being here we've got a great partnership with Maria and it comes from the work we do in sports um and the way we look at 55:39 it I was just watching a video this past week that we created around the Masters 55:45 and there's a line in it that says you see golf we see data the same thing 55:51 applies to tennis it applies to Fantasy Football it applies to all the things that we involved in and for us it's 55:58 about creating new experiences for fans to engage with the 56:03 game with whatever sport it might be in a different way for also helping the 56:08 players or the teams perform differently improve their performance on the pitch yeah and there is so much excitement 56:15 around what we do on this I I love it it's a great spot for us to I think you framed it out pretty good like how it 56:21 works for players and how it works for fans I want to dig both on both of those and Maria maybe you'd like to sort of jump in first 56:26 here is like as a as a former player like how do you view these kind of Technology products as a way to engage 56:32 more fandom yeah yeah I mean as someone that was on the tour for for more than 20 years I um you know I remember like 56:39 15 years ago my coach just before you know five minutes before I'd go in the match he he'd share with me a 30 you 56:45 know second clip of my of highlights of my opponent that would take him hours to create because he just wanted to you 56:51 know get like break point where is your server going to serve and I would see those videos um whereas now within you 56:58 know 5 seconds he has all those stats um in many ways ten is the reason you know 57:03 you go to a tennis match is for the emotional experience right but as a fan when you show up and you know everything 57:10 about the player you know everything about their matchup you even know who's the favorite how come why um it makes 57:16 that experience so much better even while you're there trying not to be on your phone and being immersed um in this 57:23 amazing experience can you bring us so you maybe spend more time watching tennis or watching tennis tournaments and playing and you used to has your Fan 57:30 Experience changed like how how are you using these kinds of tools to understand like oh I see what's happening right here in way absolutely because I I I 57:37 don't have as much time um to gather all that information I'm now a mom and I'm I'm I'm an entrepreneur and now I can 57:44 you know get that information so much quicker so when I attend an event I already know what's going on I don't need to spend hours you know figuring 57:51 out why is this opponent so much better than the other why is this player number one in the world and then one is 50 yeah 57:57 uh do you get to do that well tell us your Fan Experience we I love it I mean and like Maria is talking about for 58:04 tennis um we look at golf I've had many a golfer at the master say I go home 58:10 every night after the round and I spend all night on your app the Master's app 58:16 because you can look at every shot that every player is taking and they can all think about how do they play it the next 58:24 day knowing where the pin's going to be placed and how that changes the same thing for as a fan how do I get more how 58:31 do I get a broader experience because people don't when you think about Wimbledon and you think about the US 58:37 Open most people think about Center Court and everything that's happening there but there's 18 Courts at Wimbledon 58:43 right and there are 18 matches being played at the same time and when that that amateur who has made it in who 58:51 lives you know in Brazil and the family can't come and you can watch it online and you can engage you can get that 58:57 commentary provided wait see and engage with much more than just what's I guess on the front page of all exactly yeah I 59:03 mean seeing if you're an upand cominging player that's 17 years old and you have a fan base you know in India and Brazil 59:08 and and you get to hear commentary and all these different languages and something that you you tried out for the 59:13 first time at Wimbledon last year I remember where they did AI generated commentary for for all the courts um 59:20 which when I played that that certainly wasn't the case but but back to like the US Open and and the drawing and and the 59:27 app and where where you go to see all of that as a player when the draw came out um previously you know you know your 59:33 matchups you know who you you know who you play well against and who you didn't match up well against but now an app 59:38 tells you that so if I was still playing I'd turn on this app it basically would tell me if I could win or not which I 59:46 could use that as fire by the way the other thing that we can that we have done in the past is we look at unstructured data that's out there as 59:52 well and we've used unstructured data to make some decisions and predictions not 59:58 always public but we've made some predictions structur data is that in sports there's a certain tennis player 1:00:04 who I will leave nameless during covid who was looking at their social media feeds 1:00:11 we could see that they really wanted they love being at home just over this course of time and you start looking at 1:00:18 that thing and you see how much they're away during Co because they weren't from the US and could they go back and forth 1:00:25 and we said we think that that person mentally from what we're seeing that 1:00:30 they're publicly posting not like we're not going trolling people and publicly 1:00:35 posting and and amplifying this stuff we said we think that their brain like they 1:00:41 are they want to get home because this was the only thing that was between them and getting home and that person crashed 1:00:47 out early this this is a wild thing you're able to look at so much more data and figure it out well I'd love to hear that from like a professional standpoint 1:00:52 like from as a player as a coach like what can you learn about a player to 1:00:58 help them get better or understand what their weaknesses are like that yeah I mean you you'd almost wonder like what 1:01:03 is the position and and of a coach now right because they're all a a lot of the things once once you get on the court a 1:01:09 lot of it is instinctual but what you learn and what you process from a coach prior to the match is is very important 1:01:15 but when you get on it's very instinctual so all that data that now he can draw within minutes within seconds 1:01:22 is incredibly beneficial to me because I now know that when I'm down a break point I know where my where my 1:01:28 opponent's returning best right I know if their forehands better if their backhands better where have they done 1:01:33 you know well you know how your opponent will play because you could look at the data and it'll tell you this is what they'll probably exactly and the other 1:01:39 things is you don't need to question your coach which I occasionally did he would you know show me a few 1:01:46 videos and I was like are you sure and now he actually shows me you know valid statistics yeah I mean what you said at 1:01:52 the beginning was very interesting is that previously if You' wanted to learn about your what you would do it would take hours and hours and hours of 1:01:58 scrolling through data it would be essentially impossible now as you're walking into the I would myself spend 1:02:03 because I I I was so deeply interested in that type of data like I wanted to know of course you get out and and by by 1:02:11 being you know a player for so many years you know what it takes but if you've done the if you've done your math 1:02:16 if you've done your homework that that it's so the homework would take a long time right yes but not not not we just 1:02:23 did not with Jonathan huh not with you it's it's quick it's quick we we just did something with FC Sevilla the the 1:02:30 soccer club in Spain and we took hundreds of thousands of scouting reports that they had and we helped them 1:02:37 create an AI solution they now with natural language prompts can go and look at all these players from game reports 1:02:44 and scouting reports and understand who do they want to go sign who do they want to bring into their program even at a 1:02:50 younger age because it's great we all know Maria because of the great career she's had but as you were making your 1:02:56 run up think about how much more that this impacts the fan base as people 1:03:01 could get familiar with you so much faster and see all that that historical 1:03:06 play and understand how you play more yeah the the process of a Scouting Report is is definitely quicker I mean 1:03:12 do you how do you use it to keep an eye on people are there things that you're able to like sift through these mounds 1:03:17 of data to figure like here's somebody who has a fair specific skill set that I would have never able to figure out watch this person for a couple years I 1:03:24 mean for example there's a young girl that just won her match at the French Open she's 17 years old um she's the 1:03:29 youngest since Martina Hingis I did not know so much about her but this this has 1:03:34 been her her year she's going to be top 20 very soon and by not knowing much 1:03:40 about her I was like well where am I going to get all these stats and I went online and through AI I knew like where 1:03:46 she's performed well against what players where you know what her weaknesses are like her second Ser just 1:03:53 all these things you learn like a to cical picture you would never have gotten no so now when I go to an event 1:03:58 and I'll watch her for the first time you know perhaps at the US Open you know later this year in September I'll know 1:04:03 I'll be so much more informed about her as a player I'll go back to golf for a second what we did this year at the 1:04:08 Masters I keep talking tennis you keep talking I I'm touching them both come on 1:04:14 because it's broader I did talk soccer for a second okay yeah you did um but what we did was whole insights at the 1:04:20 Masters this year and we took eight years of data we took every shot played at the tournament for eight years and 1:04:27 you could then go look at the players who are active and say this is how they usually play this whole and understand 1:04:34 the conditions and are they more likely to perform to get a birdie or a par are 1:04:40 they playing more conservative and how does that change from Whole to whole which allowed us to be much more focused 1:04:46 and have some better predictions and out this is like a thread you can pull through almost all of these AI discussions is that it's able to boil 1:04:53 this massive ocean of data and help us understand understand what would almost essentially be impossible but I think 1:04:58 the other thing yes and the thing it allows you to also do is start adding different data because as you look at 1:05:05 all these things Maria playing tennis I would imagine that if you could lay W weather data for example on top of it or 1:05:11 you could lay what the crowd because we also analyze what the crowd noise does and how they weather like what was our 1:05:17 social media account doing you if you beyond the unstructured but if you really look at the weather and you look 1:05:22 at the crowds and you look at all those things most people would say okay they usually hit 60% forehands right but then 1:05:29 you can start looking at things in such different ways that gives you a much broader picture yeah uh we could talk 1:05:36 about sports the entire time here but I guess we got to do the business side a little bit you know it's not all fun and 1:05:41 games Jonathan is it draw this threat a little bit then for the folks in the room who maybe aren't professional tennis players or Masters level golfers 1:05:49 uh how this translates into the real world to the real world I mean what we really are trying to do is bring the 1:05:56 these are the same technologies that we bring to companies around the world I mean IBM is a true B2B company and what 1:06:04 we're bringing to the Masters to Wimbledon to US Open to ESPN Fantasy Football it's all about taking the data 1:06:12 and providing a new way to look at it and get better insights from it and get take advantage of the opportunities 1:06:18 you've got and overcome the challenges and that same application goes to the 1:06:23 clients that we work with and we bring clients through all the time at these events and they walk out and they go wait a minute can we talk to you about 1:06:30 how this is going on we never thought about it about what they were doing there I mean all these inputs that go in 1:06:36 about a tennis serve can also go into a factory floor or a financial projection it's just all numbers yeah and any like 1:06:43 you but you get your data right you can start you know looking at it and getting insights that you never had before but I 1:06:49 also think it's it's about the consistency I mean you've been partners with the US Open for over 30 years and 1:06:54 Wimbledon for that am so I think I mean that type of data in sport specifically 1:07:00 is so Niche and so to have those 30 years of data it's not like it's coming from one company to another and I mean 1:07:07 this is like an organic flow of data that's just been collected over so many years right well so like at axios we 1:07:13 like to end on one fun thing when we do AI events I like to end on one future thing so I want to spin this conversation forward right if we had 1:07:19 gone to Young Maria and said you know what in 30 seconds you could get all of these videos downloaded on a on a 1:07:24 magical phone if you gone to Young Jonathan and said you can predict all of these Master 1:07:30 shots you'd be like that's crazy talk where is this going like what's Over the Horizon that gets you really 1:07:35 excited I mean the the the experience of attending event is just you're so you're 1:07:40 so well informed like when I used to go to you know I didn't have so much time when I was on tour for a full you know 1:07:46 10 to 12 months of the year but when I would go to another sporting event I didn't I really didn't know much I was 1:07:52 just there for two hours just figuring out like how good is this Guy what's you know who's going to win why how come and 1:07:58 now I just I feel like myed knowing what experience is so much better because of 1:08:04 it and I know it's it's it's all numbers and it's very scary but it's very real and it makes us I think experiences are 1:08:11 I mean we can buy a lot of things and you know appreciate them for a little bit of time but experiences are forever 1:08:16 and if our experience can be that much better I mean isn't being there to make it that much more Rich right is so 1:08:22 fruitful all right Jonathan take us into the secret IBM Labs what's cooking I what's cooking is in these spaces are 1:08:28 things that we're going to work on with the US Open with Wimbledon with masters with ESPN moving forward and we've got 1:08:36 some interesting stuff that you know Wimbledon will announce in the not too distant future in the US Open tell us 1:08:41 now yeah yeah almost got me um trying I tried backstage tell me to your point 1:08:47 Maria it's the experience what we're what I will tell you what we're going to do it's about the experience it's about the experience of the 15,000 fans who 1:08:55 are in in the stadium but it's also about the experience of the millions of fans that are watching online and how 1:09:02 they can better engage with it and have a better understanding of what's happening so you've got the online but 1:09:08 then the other thing is you can take that and change that experience so when you do go for two two hours to an event 1:09:16 you understand the data of what's happening there but you also can get a better experience as you go through the 1:09:22 grounds right you know that you should be going to watch this match here you know that if you want to get something 1:09:28 you know the wayfinding that can come with these sorts of things like the Journey of exper like of attending an 1:09:34 event will start to benefit from this think about going to a stadium and you get up and you're like okay I've got 1:09:40 five minutes between a quarter of basketball or 20 minutes for hockey or whatever it is I want to go get a beer I 1:09:45 want to go to the bathroom very simply if you could open your app and start saying and it's knows where you are and 1:09:52 it says go left cuz the line to the bathroom is shorter and go or go right if you want a beer oh I need that 1:09:58 exactly but you think about that cuz it also maximizes when I would step out at 1:10:03 Wimbledon last year like you know what it's like Maria they keep you out of there until like know it's like 1:10:09 Wimbledon right I actually don't I'm going to be no I genuin I'm going on the 1:10:16 grounds this year for the first time I'm going to like do a whole walkthrough of the grounds which I haven't done in many years so so fig figuring out other ways 1:10:22 to to take all this data and interpret it and provide Ed to the person watching attending it to have the best experience 1:10:28 possible virtually or in person that's what's going to continue to happen and it is massive the opportunities perfect 1:10:34 uh fun things to look forward to Maria Jonathan thank you so much for being here thanks for IBM for making today possible stick around the conversation 1:10:40 continues thanks thank you 1:10:46 [Music] [Applause] 1:10:57 [Music] 1:11:14 please welcome axio Pro Tech policy reporter Maria curri 1:11:23 [Applause] hello everyone it's great to be here on the AI plus stage with you today you 1:11:30 know these days when we think about content creators we often think about influencers and Tik Tok but there is an 1:11:37 entire ecosystem out there of content creators whose livelihoods are really 1:11:42 being impacted by Ai and that's why I'm so excited for this conversation that we have coming up here we're going to hear 1:11:48 from Michael Sanchez the CEO of rapti an ad management company that helps 1:11:53 creators build independent comp companies and Brands and we he will be joined by Lisa Brian who is the founder 1:12:00 and creator of downshiftology a recipe site and she will attest to how AI has helped her 1:12:05 grow but to also how it could obliterate the Creator's online economy entirely 1:12:11 Michael and Lisa come on out 1:12:17 hello welcome 1:12:22 hello okay so we're going to get get a slide up here now and kind of set the 1:12:29 scene with some numbers Michael if you can walk us through what rap is and the impact that it has thanks Maria for 1:12:36 having us hi everybody I'm Michael Sanchez CEO of rap of delighted to be here uh our mission at rapv is to help 1:12:43 the highest quality content creators build sustainable businesses and live their dreams we now serve 5200 1:12:50 extraordinary entrepreneurs across every state in the US you'll meet Lisa a great example short shortly um together theyve 1:12:57 produced 25 million pieces of copyrighted content um really amazing 1:13:02 when you start to look at these creators they're doing this full-time they're making a living they're feeding their 1:13:09 families they're creating tens and hundreds of thousands of jobs that's where rapv comes in um we say hey Lisa 1:13:16 and creators focus on the content we handle all the advertising and their monetization help with different parts 1:13:22 of their business um this has been an amazing journey and I think it's one of the kind of Secrets of the internet and 1:13:28 the open internet is the scale that these extraordinary content creators have reached together the rap of group 1:13:34 of 5200 creators now reaches 191 million monthly visitors in the US this is each 1:13:41 month we're now at top 10 I think number eight comcore property scale that's now Beyond many of the big platforms and the 1:13:48 media companies we all know this has resulted in rapv uh becoming the largest 1:13:53 source of ads on the open internet I think there's good data out there that 15% of every dollar spent on the open 1:13:59 internet now goes to a rap of Creator which is extraordinary $2.5 billion um 1:14:05 have been paid to to creators and I think we're especially proud of the fact that 58% of all diverse owned media 1:14:13 businesses are with rapv and um that is really a growing number that's very important for for the health of the open 1:14:18 internet I wanted to focus on that number for a little longer why is it important to have diversity in the 1:14:23 Creator community good question th think of the promise of the internet Not only 1:14:29 would billions of people be able to get extraordinary content but it could come from all different diverse voices great 1:14:35 perspectives information entertainment um just within rapti 80% of our creators 1:14:43 our entrepreneurs are women-owned female businesses and we have a rapidly growing number of blackowned Hispanic Asian and 1:14:49 these are the diverse voices that give the internet important diversity and 1:14:55 we're going to obviously we going to talk about AI here but if you abstract out all of that personality you're just going to have a plain vanilla answer 1:15:02 we're going to lose the diversity um of what's uh of what makes the internet special and even when you look at 1:15:07 results we're going to lose this very important part of of the open web another figure that really strikes me here is the 25 million copyrighted works 1:15:15 that have been published online that's just an enormous number talk about the importance of that of that work think 1:15:22 about the Blood Sweat and Tears and the creativity and the Brilliance of the thousands of creators I've got a few 1:15:28 thousand of those on my site I'll take credit for those yeah thousands and thousands pieces of content that are 1:15:33 express its narrative its photos its images its stories all copyrightable all 1:15:39 protected and now as we look towards um generative AI we're very concerned that all that gets lost in the desire to put 1:15:46 one answer forward into into a new experience and we'll definitely get to that if we can queue up the second slide 1:15:52 here and kind of get into why all of this matters and turn it over to Lisa um Lisa tell us about what downshiftology 1:16:00 is um and how has it become your livelihood sure so about 10 years ago I started down shiftology I say in a blaze 1:16:07 of burnout from the corporate world where I was probably working too much had a little too much stress and wasn't 1:16:12 eating healthy and I was going through some health issues at the time and so I left the corporate world and I started 1:16:18 this little website where I thought to myself I'm just going to share some healthy recipes and hopefully help some 1:16:23 others out there who might want to downshift in their lives as well and over time what started at this as this 1:16:30 little passion project turned into a vibrant digital media company where my 1:16:36 audience grew over time and the entire ecosystem of independent Publishers grew as well and now today downshiftology 1:16:43 reaches over a 100 million people a year which is just mindboggling yeah 1:16:48 absolutely um and then just to zoom out you know this is uh 78% of creators 1:16:54 their number one concern is that AI will negatively impact their income can you 1:16:59 zoom out a little bit and tell me about what you're hearing from the other creators at rap of supports maybe just to put all of this in context and uh the 1:17:07 AI platforms open Ai and Google which we're most concerned about at this moment they have scraped the whole 1:17:12 internet the 25 million pieces of content have all been scraped that's been their training data and I'm sure 1:17:18 everyone saw their announcement a couple weeks ago but they are in full roll out of changing the search experience now 1:17:25 called AI overviews where they've trained on leisa's content but now the results will be taken from derived from 1:17:32 leisa's content and all the other content creators and we'll give an AI answer and then maybe some links we'll 1:17:37 talk about that I'm sure but the real danger here is that their content's been trained upon and then the traffic the 1:17:43 lifeblood the revenue that goes to all these content creators will go away there will no longer be the traffic back 1:17:49 Revenue will be hurt and so when you talk about 78% of content creators are worry about AI let's put a human element 1:17:55 on that of of the 5200 creators we serve many of them are making 675,000 a year 1:18:01 if you were to cut their income in half which is a very possible scenario as this rolls out that's devastating to 1:18:06 their livelihoods for the larger creators they are um uh hire they're employing quite a few employees um 1:18:13 there's no wonder that creators are extremely worried about their livelihoods I think that's a really important point because a lot of like a 1:18:19 lot of people out there see us as solopreneurs independent you know creators but it takes a team to create 1:18:27 high quality content and so as creators start to lose traffic to their sites and lose their livelihood it's not just 1:18:33 potentially creators going out of business it's all of the people that they employ as well and so I think that's a really big concern and then to 1:18:39 the AI overviews Point you're getting these results that I call it Amalgamated mish mash you know it's taking little 1:18:46 bits and pieces and stripping out the humanity stripping out the personality and the perspectives and the unique 1:18:51 viewpoints and so you're just getting this very homogenous content without what makes each of that 1:18:59 those unique pieces of content you know valuable so we've talked about the revenue concerns the concern that you 1:19:04 know your employees might also be on the line here if we go on to the third and final slide we'll see a layout of the 1:19:11 different concerns if we could get to the third slide please so let's talk a little bit more 1:19:18 on the misuse and competition aspects of of this entire story what are the 1:19:23 concerns around misuse Michael we still see it every day and there's a lot of examples of AI 1:19:29 misfiring but they are putting out answers in content sometimes even citing our creators that in the the silly 1:19:36 examples hey here's a blueberry pie recipe and then all of a sudden they're cting Creator and the recipe is terrible 1:19:42 um and so there's real concern and I don't know leis if you want to talk about competition I I think I kin it to the one of the biggest heists of 1:19:48 copyright infringement happening right now and if I were to go into chat GPT or Google Search and say so I have a 1:19:54 published Cookbook by penguin random house and if I were to say hey give me paragraph one of Chapter 3 of my 1:20:00 cookbook or if I were to say or if on YouTube I were to put a 3 second snippet 1:20:05 of a Taylor Swift song both of those things would get swiped down immediately for copyright infringement but on the 1:20:11 open web right now our content is just being scraped and repurposed and that 1:20:16 copyright is not being acknowledged at all and are you worried that this will increase competition against you know 1:20:22 your work abs absolutely what it's happening is it's increasing competition against myself of my own work so it's 1:20:30 scraping my content and my recipes creating a mishmash form of that and then competing head-to-head with me so 1:20:37 so we talked about um the announcement that Google recently made that it now has these AI generated summaries when 1:20:43 you put in a query um I want to give you some of the arguments that Google um has 1:20:48 made for why this is actually a good thing and then have you respond so Google says it's actually showing more l 1:20:54 links with its AI overview feature and that those are getting more clicks which is ultimately beneficial for Publishers 1:21:01 what's your response Michael we hope that turns out to be true um they were testing for the better part of a year 1:21:08 and we had very deep analytics on all their testing when they would test using the search experience certainly in the 1:21:13 testing we saw substantially fewer clicks and then when you look at the actual experience that they're putting 1:21:19 forward here is a whole generated answer and maybe here's a link or two we don't 1:21:24 think that be more traffic um the number one thing we're asking for and you know the the CEO of Google was out there 1:21:29 publicly in interviews being pressured and the question is just be transparent show us the amount of traffic that's 1:21:36 going to sites and if we could look at that there maybe is a happy result but it seems pretty clear that the 1:21:41 experience as designed will mean far fewer clicks and back to the diversity 1:21:46 it's going to be a small number of big players who are getting the few clicks that are left so we're very concerned transparency would be one part of the 1:21:53 solution Alisa would you have anything that yeah I would say what I see which is a little scary at this point in time 1:21:59 is this fractured internet where you're going to get AI 1:22:05 overviews that again are these homogenized answers from a variety of sources and then content creators are 1:22:12 either going to get put out of business which we've already seen and this is everything from people in food and 1:22:17 travel and gardening I mean it spans so many things you know just think of all the queries that we all do on a regular basis and so then the content creators 1:22:25 that are left are most likely going to start pay Walling their content to protect it and so instead of a free and 1:22:30 open web that benefits everyone you're going to end up with homogenized content or pay Walt content and I think that's 1:22:36 just a sad future to Envision another argument that Google has made is that you know the AI summaries will lead to 1:22:42 growth for high quality content um I think we all have used Google in our lives and sometimes we'll search 1:22:49 something and you know we don't get the answer we want right away what would your counterargument be to this being 1:22:54 ful for consumers and having an immediate better response at this point it is not 1:23:00 extremely useful I think over time to be clear AI will get better and better it become more useful I think a big part of 1:23:06 what we're saying is it is the content creators who created all the content that has been trained upon and they 1:23:12 shouldn't be left behind they should be a commitment that traffic is going they should be compensated for being part of 1:23:19 all this value creation so the you know there was all the news out there of all the silly results uh last week but over 1:23:25 time it'll get better you just can't leave content creators behind and have their livelihoods go away when it was their content form the basis of all of 1:23:31 this and Google has recently had to roll back their search function a bit after people were getting results from you 1:23:38 know suggestions to eat rocks or put glue on their Pizza um is there a bigger 1:23:43 danger here to having an ecosystem of information that is unreliable Beyond these sillier you know cases absolutely 1:23:51 I mean so my website alone is gluten-free recipes um there are a lot of people out there with autoimmune conditions if you know if a result gets 1:23:59 returned say somebody wants to swap an ingredient and Google gives them a wrong or incorrect response I mean that could 1:24:04 be detrimental to a lot of people out there so I think these answers that are unverified and taken from a variety of 1:24:11 sources whether it be Reddit whether it be chat groups where you know it's really hard to differentiate what's true 1:24:16 what's been said in just I think going back to the original content creator who has the expertise in that area and it's 1:24:24 there lived Human Experience that has given them the authority to create that content I think it needs to continue to 1:24:30 go back to the original content creator maybe Maria going back to your question about whether more traffic is going to go out to the open internet um we all 1:24:37 watched their presentation um announcing rolling out open AI uh uh Ai overviews 1:24:44 and it felt very clear what the vision was coming from Google and it was you know we'll do the Googling let Google do 1:24:50 the Googling for you so it used to be a user going to search clicking into a 1:24:55 site experience it and going out what they're saying is it's all going to live in their wall Garden you don't need to 1:25:00 go out so I think the intention is very clear certainly raises anti-competitive things money used to revenue used to go 1:25:07 to creators now we live on those walls so it didn't feel like their intention as the product was described let's see 1:25:13 but let's make sure they're accountable to make creators whole and to be transparent about what's happening but AI isn't going anywhere right and so you 1:25:20 touched on this a little bit Michael already I'm wondering if you can expand on it what do you want from Google what do creators want I I think what's most 1:25:28 important is that creators participate in the value I think first and foremost if we're ticking through step one be 1:25:34 transparent with what's happening in the results that will be very helpful Congress who we've been spending some 1:25:40 time uh talking uh to the legislators and staff on both sides of the aisle typically the Senate Judiciary Committee 1:25:47 both sides are very concerned about anti-competitive and copyright and fair use but a clear statement from Congress 1:25:53 saying uh this is copyright infringement and farious doesn't apply would be great pressure on Google uh law enforcement 1:26:00 agencies the Department of Justice and FTC they could get in they can get involved and I don't know if we're going 1:26:05 to talk about this but creators are starting to mobilize and and and they're um really looking for Google to say make 1:26:12 sure you know the commitment from Google Maria could be creators will either not lose traffic or we will compensate them 1:26:18 for what's lost they should be participating in whatever great value is developed and I think that there's a a big movement right now with content 1:26:24 creators educating their audiences and saying you know in the future this is a problem we are seeing you know can you 1:26:30 tell the difference between an AI generated recipe and an authentic recipe and I did that with my audience on 1:26:36 Instagram last week and they could not tell the difference between a real and an not and if you're going to spend $50 on expensive ingredients because you 1:26:42 want to make a filet Manan you want that to turn out right you want to know that it has been tested and tasted and last 1:26:49 time I checked AI could not test and taste a recipe and I think it's just like food more than any industry is you 1:26:55 know the core of the humanity you know we have to taste it and we have to you know my audience follows me because of 1:27:01 my specific taste and so I think yeah we're just going to lose all of that and 1:27:06 what are the pressure points here is it going to take litigation is it going to take a change in laws the litigation question is very 1:27:13 interesting um The New York Times is out there sing suing open AI but it doesn't appear that anyone has challenged Google 1:27:20 and they're certainly very well known for taking many many years in defending very aggressively so smaller companies 1:27:26 and entities pursuing that that could take years and years and maybe somebody doesn't I think that would be is that 1:27:31 something that you would pursue I I think companies like rapv and other media companies are exploring all options as as as we will continue but 1:27:38 it's very daunting to to do so and I think the more angles of pressure can put on Google hopefully they come to the 1:27:44 conclusion that letting uh and it's good for the internet if all these creators go away what gets trained on it's going 1:27:50 to be this plain answer there so we we really hope that the creators and these websites don't get abstracted away talk 1:27:57 to us a little bit more about what your playbook in DC has been which lawmakers have you been meeting with how have they 1:28:03 received your message and can we anticipate any action on Capitol Hill when it comes to this you know especially in an election year I I I 1:28:10 think you all are more expert on what Congress will do it feels very hard to imagine they're going to pass 1:28:15 legislation but both sides of the aisle we've been spending time primarily in Senate Judiciary Committee on both sides 1:28:21 they're very concerned about the anti competitive side of what Google has done just one example um they're now saying 1:28:29 hey you could block creators can block the crawler you will then not show up in the results for open for for the AI 1:28:36 overviews this is something that's anti-competitive so I think a lot of people on both sides are looking the the 1:28:41 anti-competitive side and maybe there's an angle there that that that that gets pushed through and have you felt that 1:28:46 there's a bit of a gap on the house side of things I mean do you have anyone there that's willing to sit down and talk about these issues we have started 1:28:53 on that side again it is knowing the US legislative system it is not something 1:28:59 that you can put a lot of hope that they're going to pass stuff but they put pressure their letters their hearings and again hopefully and you know Google 1:29:05 is saying a lot of the right things we hope they end up doing a lot of the right things all right well that is all 1:29:11 the time we have thank you so much Michael and Lisa for being here everyone please stay tuned for two more segments and then we'll take a short break thank 1:29:17 you thank you good job thank you 1:29:24 [Music] 1:29:49 [Music] he 1:29:56 [Music] 1:30:04 returning to the stage Ena [Music] 1:30:12 freed so you may or may not have heard of the Allen Institute but they are a very important player in the AI research 1:30:20 field they've been a big advocate for making powerful open ource models available so that researchers can do the 1:30:27 work that they need to do to understand this technology and as you've heard already and you'll hear more of we 1:30:33 actually don't understand exactly how the technology is doing what it's doing 1:30:38 now it's not new to technology that we may not understand what it's doing but 1:30:44 it's quite novel that the people building it can't even necessarily explain what it's doing um the Allen 1:30:50 Institute earlier this year released an open-source large language model uh which our next guest said will 1:30:56 fundamentally change how researchers and developers learn and learn about and build AI please join me in welcoming the 1:31:03 CEO of the Allen Institute for AI Oli farhadi 1:31:09 Ali thank you now Oli you're in a fairly new 1:31:14 position heading the Allen Institute but you're not just a professor not just a professor professors are wonderful but 1:31:21 you're also a tech entrepreneur you were at Apple um all of that I think 1:31:28 experience is really useful when it comes to this moment um I want to talk about a bunch of things but one of the 1:31:35 things I think that is good to address and I'm really glad is you know there's a lot of debate about open source closed 1:31:41 Source often in the te context with AI of one is inherently safer or better so 1:31:48 the critics of Open Source say oh you're putting the most powerful technology out there it's easy to remove the safeguards 1:31:55 um it can be used by Bad actors um what is the case for open source when it 1:32:02 comes to Ai and generative in particular uh thank you I think I'm happy that you're asking about openness 1:32:09 um it's an overloaded term these days um I think first maybe I actually start talking about the notion that's been 1:32:17 implicit on the streets that close and safe are synonyms to me we are probably for the 1:32:26 first time are deploying a piece of technology that we know very little about and many 1:32:33 of the things that we discussed today about this this Behavior versus that behavior all of them are actually rooted 1:32:40 and the fact that there are major technology gaps in these models that 1:32:45 we're dealing with these products that we're actually shipping at the scale and 1:32:51 AI excluding the the last few few years was born and raised in open we are here 1:32:56 in AI today because AI was practiced in an absolute open terminologist I built something in an 1:33:03 open I released everything you build on top of mine and then one step after the other till we get to the point that we 1:33:09 got today it wasn't over time it wasn't overnight it wasn't by one group it was 1:33:15 actually communal decade long effort to gets us to the point that where we today 1:33:20 um now closing a piece of especially in mature piece of technology early on is 1:33:26 probably one of the biggest standard that we're worried about we don't know enough about this 1:33:32 Technologies and we're depriving the the talent the brain talent that exists in 1:33:38 the industry in Academia in research labs in startups who can actually contribute to close those technology 1:33:45 gaps by keeping the technology behind the closed doors and that to me is just much bigger threat to an existential 1:33:52 threat to me in in my opinion to AI than other hypothetical scenarios that we're talking about it absolutely like any 1:33:59 other piece of technology technology can be misused but now we get to decide which 1:34:04 word we would like to live in a a word in which the technology will be misused and we're now facing with a hypothetical 1:34:11 threat but only a handful of people know how to fix it or award in which we actually have millions of experts 1:34:17 millions of players who can actually jump on the problem and solve solve the problem um but but aside from open and 1:34:25 close open by itself is an overloaded term I think there's a whole spectrum of 1:34:31 openness when we think about openness we have on this side of the spectrum we have absolutely closed behind a API 1:34:38 models we don't know much about them uh we can query them and get response back if I move a little towards openness we 1:34:45 have open weights models and we've seen various versions of those I think those are great initiatives in the right 1:34:51 direction so to help people out open a I that's an example of you just have access to the API you're not really 1:34:57 seeing behind the curtain something like what meta's done with llama llama 3 you have open weights but not necessarily 1:35:04 why is that not everything yeah so these are not open models these are models 1:35:09 that are trained behind closed doors the data that was fit into these models is notknown the training algorithms the 1:35:15 recipes all the all those elements are unknown but the model weights are being 1:35:21 shared which is great we appreciate it we cherish that was celebrated but it's just not enough for what we need the 1:35:27 most important element as we actually we as a community learn about these models the most important element of these 1:35:32 models is the training youa that went to this models and that's what you all have done that's different so earlier this year um the Allen Institute released a 1:35:40 large language model not just with the weights but with the training data what has that enabled already what's been the 1:35:47 response yeah we've been actually surprised by the impact that dolma or 1:35:52 open data set and all more open language open language model framework has empowered people have used these kind of 1:35:59 Technologies beyond beyond the ways that we imagin they would it has actually shaped PE the way people understand 1:36:05 these models to some extent empowered people to research about the relationship between input and output of 1:36:11 these models and this is exactly the kind of behavior that you would actually get from an open initiative you impower 1:36:18 others to do their research you Empower experts who were sitting idle now they could actually start building up on 1:36:23 these Technologies if I only release my weights it's great others can actually use it and find tunity to their task but 1:36:29 that's is not barely enough for Innovation to Gap to to feel the Innovation Gap that exists researchers 1:36:35 innovators they actually need to need to have access to the whole stack a key element in open open source is the 1:36:42 ability to build on top of what you've built and what you guys have released is obviously very important it allows some 1:36:49 of that research that you talk about but also the research Community needs to understand what's going on broadly and 1:36:56 that requires a lot of computing resources whether the model is open source or not do we need some sort of 1:37:02 I've heard people uh talk about the need for some sort of national compute resource or something where researchers 1:37:09 can not only test but really use at at scale some of these models both yours 1:37:14 and others uh absolutely I think there are two answers to that problem one is that by sharing by opening up the whole 1:37:21 stack you're actually empowering others to grab your solution halfway and build 1:37:26 on top of it Fork from it in a different way and this means that you've already saved this much of GPU also you you you 1:37:35 we spend a lot of GPU hours trying to reverse engineer What You've Done Right you've built something great shiny put 1:37:41 out there I say I want one of those you're not going to tell me so what do I do I just try to reverse engineer it and I'm going to repeat the same mistakes 1:37:47 that you have done and just globally if you look at the carbon footprint of that the energy waste the water evaporation 1:37:54 Wass this is just not going to serve anybody but aside from that I think we do need National initiatives from that 1:38:00 and President Biden actually started near the national AI resource um and I'm actually very happy we're co-chairing it 1:38:06 we're very happy to see how it's evolving the first set of Grants are out uh we need way more of that we need a 1:38:12 lot more resources to be able to empower the talent that already exists in this country to start filling those those 1:38:18 technology gaps many of the problems that we think about today in terms of policy regulation yes they look like 1:38:24 regulation problem but as a matter of fact there are technology problems the root cause is actually the inability for 1:38:30 us to control the output space of these models let's say parametrically and all of those will actually those 1:38:36 conversations become much easier when we mature up the technology what is the best way to mature up technology the 1:38:41 most effective way open it up Empower every everyone every players in the market to actually start doing innovating in that space and things like 1:38:49 National AI National AI resource is actually a key step in into that step 1:38:54 but not barely enough we need to do 10x if not 100x of that so the openness is one issue Computing resources is another 1:39:02 um it seems to me another and you've talked about this some is this issue that um you know we we went from you 1:39:11 know technology that was being developed in the open in researchers to an 1:39:16 explosion of availability I think how many people in this room use generative AI at least once a week 1:39:24 so the good news is everyone has access but today's Genera of AI has a lot of 1:39:30 flaws and a lot of problems talk about what does it mean that we have such a powerful technology I think you've 1:39:37 called it an immature technology being deployed at scale what does that mean and what does that mean for 1:39:43 society so I think when when you think about technology and Society I I always 1:39:49 Envision a curve where the technology becomes so WID spread and so popular 1:39:56 that is out of your face internet electricity all of those 1:40:02 are examples that when we're dealing with them we don't realize we're dealing with technology we just take them for 1:40:08 granted my daughter such a hard such a she has such a hard time believe 1:40:13 understanding why she cannot watch YouTube while we were in playe we just take it for granted internet is always 1:40:18 there so I think a technology would actually need to go over that hump and 1:40:23 becomes an ambient part but that happens only and only if we can actually 1:40:29 establish trust between those between the technology in humans and that trust right now is broken in my opinion and 1:40:35 needs to be earned back it's broken because blackbox nature of the practice is really hard to get to earn the trust 1:40:41 of a people given the opacity of this whole practice also the trust is broken 1:40:46 because these models are not 100% reliable they hallucinate they get out of Norm we don't know how to control 1:40:53 them and we need to hack around them um also we can't the trust is broken 1:40:59 because the evaluation is broken we actually are in such an evaluation crisis today and I wanted to talk about 1:41:05 that because you know early on some of the things that really got people excited about generative AI was you know 1:41:12 oh it can pass the bar it can pass the medical school exam and then it was you know months or you know a year later we 1:41:18 realized oh it just really understands how the test is written um and that's not not just limited to that we have a 1:41:25 hard time saying how good is this model how good is that model how good are they at a particular task that's something 1:41:32 you're working on what are you doing yeah so I think we are in the evaluation crisis mainly 1:41:38 because our the the capability of these models the r that or the pace of these capabilities or us discovering those 1:41:44 capabilities severely outpaced our ability to evaluate in AI it was the 1:41:50 other way around before we build a piece of technology actually before before we work on Ani task we thought about an 1:41:55 evaluation we put an evaluation in place we put metric in place and we start doing practices and there were protocols 1:42:01 around it here's my train set here's my test set when you're training an algorithm make sure the test set doesn't 1:42:07 leak into the train set and then we actually had metrics to evaluate those but given how broad broad these 1:42:14 Technologies the new new recent llms and generative AI in generals are and how WID spread they are we are not in that 1:42:22 luxury anymore we're now in a situation that we discover a new capability that we haven't thought about it before and 1:42:28 now we're scratching our head how can I actually evaluate these Technologies and now we're also given that we don't know 1:42:33 what was fit to these models we're scratching our head was this part of the training data was if this was not was 1:42:39 something very very similar to this part of the training data and without actually openness it is hard to start 1:42:45 being scientific about the evaluation and that's come up in a lot of ways one of the ways it plays out is all these 1:42:51 intellectual property lawsuits because you know open Ai and Google and meta 1:42:56 aren't even telling us what's in the training data it sure seems to a lot of creators of art and literature and music 1:43:04 and videos that their data must have been in the training set but we don't know that's one set of concerns 1:43:11 attribution and compensation for creators but also accuracy who's 1:43:18 deciding how important is it to solve some of the problems of generative a AI 1:43:23 before we just leap to ever more powerful models um it is absolutely essential in both internally and 1:43:29 externally as an internal internally as a developer of these Technologies I need to compare my this month of effort 1:43:36 versus last month did we make a progress or not and we need to establish some some some Proxes all of the players in 1:43:42 this space have internal Proxes for those many of them are they even question their own Proxes but they are 1:43:47 there as a as better than nothing U but the other issue that we have with evaluation specifically evaluating 1:43:54 things that are so WID spread is the access for which we need to evaluat are 1:43:59 unknown or we will be surprised around it and I think similar to the initiatives that we had around national 1:44:07 resources for for to empower others there should also be national standards for these and I'm actually sort of super 1:44:13 excited about what nist is doing in this space we're trying to National standards body uh and we're trying to actually 1:44:20 collaborate with them to figure out what's what is the right way to think about this problem what are is there a 1:44:26 hope for standardization you see tables coming after tables every week and we 1:44:31 sort of at ai2 we actually created some of those benchmarks but we look at the way people actually compare results over 1:44:37 that Benchmark it's laughable it's just the last thing you could call a scientific approach and the issue is 1:44:44 that Protocols are not the same people are comparing apples and oranges no one talks about data set leak people don't 1:44:50 talk about is this a right metric or not there are just a subset of data sets that are out there and I pick and choose 1:44:56 whatever that I like and report on them also if I know you're going to test me on those data sets what do I do I just 1:45:03 develop expertise for those benchmarks and we've seen claims after claim that my my my model better than your model on 1:45:10 those six things but the minute I actually try it on the seven thing Things Fall Apart given how much is at 1:45:16 stake in terms of these tables and claiming mine is better than yours I think we do believe we need National 1:45:22 initiative in figuring out what it means for a Model A to be better than b and evaluating is just saying how good is it 1:45:30 at X or Y or Z how important is also the next step understanding how the model 1:45:35 came to the conclusion it did which so far we've basically built most of the generative AI landscape without really 1:45:43 understanding how they work there is some research anthropic just had a really interesting paper of 1:45:48 actually understanding how it reached the conclusion without having to fully comprehend a billion parameters um I'm 1:45:56 curious is that a worthy goal is it important and How likely do you think we 1:46:01 are to continue to get to understand how today's models works let alone 1:46:08 understand how the next generation of models works I think it's an absolutely essential thing I think going back to 1:46:13 the trust trust is broken we have to proactively try to earn that trust 1:46:18 especially for regulated Industries for sensitive Industries for industries that matters you actually be able to explain 1:46:23 your way out of a decision it is absolutely critical for you to be able to Anchor a decision a generation into 1:46:31 grounded facts whether being retrieved set of evidence retriev documents being explanations of neuron activations or 1:46:38 features which are just linear combinations or nonlinear combinations of those features every every initiative 1:46:43 in that regard I think we we highly appreciate it we're far from being able to explain our way out of those but 1:46:49 we're working on it and a key piece to that goes back to open data the mean that you actually are willing to open 1:46:54 your training data there are things that you can actually ground your response at and without it we have to explain our 1:47:00 way in weird ways so we only have about a minute left I'm curious if you could put more resources toward one thing for 1:47:08 society to better prepare itself for this moment what aren't we investing enough in yeah I think um 1:47:16 educating a broad range of society so our society consists of users expert 1:47:22 users novice users decision makers policy makers and I think one piece that's missing is being able to explain 1:47:30 the complexities around these models in a way that connects to the masses and going back to again the earning the 1:47:36 trust back I think the minute that people understand what's going on inside these models the minute that we 1:47:43 understand what's going on Within These models we would have a different conversation around these Technologies 1:47:49 and if I want to spend a dollar on this I would actually spend it on understanding what is it that you actually deploy your 1:47:55 skill now prior to uh taking your current job as head of the Allen 1:48:00 Institute you were at Apple we haven't heard a lot from Apple and AI they're talking about you on Monday how excited 1:48:06 are you I'm excited coming on Monday yeah I'm excited to see what's cooking on DDC are we going to see some of your 1:48:11 former work you think um do you want to put me in apple jail all right well we'll find out on 1:48:17 Monday we'll both be watching thank you so much thank you so much great meeting here great to see you all 1:48:24 [Music] 1:48:34 up next axio senior business reporter hope 1:48:40 King all right hi everyone I know I know you really want 1:48:46 to take a break and I promise you will get the full break but I need my time 1:48:52 with my denim jacket up here I got permission just FYI to um it's so great 1:48:57 to be with you guys here in New York talking about Ai and all the changes um we've been covering music in depth here 1:49:03 at axio and the relationship between Ai and music back in November at our AI 1:49:08 Summit in DC I spoke with the lead member of Fitz and the Tantrums as well as the CEO of Sound Exchange and in that 1:49:15 conversation we focused on licensing issues we focused on how AI is impacting the creation of music but in this next 1:49:23 segment we're talking about how AI is impacting the listening experience which is most of us unless some of you are 1:49:31 secret rappers and musicians here when I'm sure there are ton of um and so very specifically I'm talking about spotify's 1:49:37 Aid DJ and for hundreds and millions of you who are Spotify users you probably 1:49:43 already know what I'm talking about but for those of you who are not as familiar 1:49:49 I want to please introduce you to X thanks for the intro hope hey what's up 1:49:54 everybody it's me Xavier my friends call me X I think you know me I'm spotify's AI DJ I'm super excited to be here at 1:50:02 the axio AI Summit you know I love anything Ai and hey humans are all right too especially this human right here 1:50:09 it's the real X the real X yes 1:50:14 Xavier so good to see you so we're going to we're going to try to I mean I feel 1:50:20 like we need a little can I do New York City Make Some Noise what's 1:50:25 going on there you go all right I got one more thing I normally do okay my 1:50:30 name is Xavier my friends call me ex before I host anything y'all got to help make me feel welcome help Hope feel 1:50:37 welcome say what up X you got to do way better than that act like you at a concert or something say 1:50:42 what up x what up X we ready I want you as my co-host for every single one of 1:50:48 you saw me struggling up here with these guys a little bit right no I wasn't I was great no you you killed it thank you 1:50:53 yeah I was watching so so you are a voice model so you're not just a pretty voice you actually do a lot of writing 1:51:00 talk about the process of actually creating the DJ but also the ongoing work that you do yeah the initial 1:51:06 creation it's like a recording process but what you hear like what you just heard aren't actual recordings they're 1:51:13 The Voice model it's text to speech and the way we train The Voice model it's like a recording session so I'm voice 1:51:19 acting from a script and I have to be very accurate because we're training a voice model what does a period sound 1:51:25 like what does a comma sound like a slight pause but I'm voice acting as myself so that's how we train The Voice 1:51:31 model and it's repetition it's scripts like you hear a ax say when I hear the 1:51:40 AI talking to me the pauses sound so natural how do you make those types of slight indentations and stuff like sound 1:51:47 like you're talking to me and not computers trying to generate something that is talking to me with the 1:51:52 technology we have it's all about the performance what you put in is what's going to come out so for me it's all 1:51:58 about being authentic being natural to who I am the word choices so that writer's room that you alluded to part 1:52:04 of it is it's what you hear but it's also what I'm saying so there's a writers room that we have is generative 1:52:10 AI but we also have our team of Music experts the foremost musical experts and 1:52:15 culture experts on the planet are in this writer's room I'm in the writer's room I'm one of those people too our 1:52:21 data experts are are ux writer so all of us are in this writer room and it's all 1:52:26 about using words that I actually say so you'll never hear AI DJ say something or 1:52:32 AI ex say something that the real X wouldn't actually say and those words include Tunes so X will never say the 1:52:38 word tun what are other words that X will never say because you don't say in life I tell you words how about I tell you words I will say instead of what I 1:52:44 don't say CU I don't usually think walk around thinking about what I want but I do know you walk around with a notebook sometimes to kind of jot down some of 1:52:50 your thoughts so so I don't say Tunes but say jams Bops bangers hits you know what I mean so things like that yeah so 1:52:56 like when we you press the DJ button it goes back right I'll come back if you like want to change up the vibe I'll say 1:53:03 okay I got you so I say I got you a lot or I'll say okay Bet coming right back to this those types of things or when it 1:53:09 comes on in the morning I say yo what's going on hey what's going on so all those different ways so I I took a 1:53:15 notebook around about like using phrases that like I actually say in my day-to-day life I think the other really 1:53:21 fascinating part of your job is that you are in there updating x with real current event knowledge it's it's not 1:53:27 just one and done how often are you there updating it's a weekly writers room so it's like what's happening in 1:53:34 the Ze guys think about it we as Spotify we're talking to the actual artists we're listening to the albums early we 1:53:39 know what's going on in culture we're talking to the labels so we're putting all of that information right into the 1:53:45 experience each and every here's an example okay the Kendrick Lamar Drake beef do we know about this y'all know about that yeah yeah y'all know about 1:53:51 that right how would AI know about that we as the humans have to put that in so 1:53:56 the human element is key to what we're doing with AI DJ and I think the other really important thing is that you have 1:54:03 this massive scale yeah hundreds of millions of people across the world are tuning in they have this relationship 1:54:09 with your voice now and that's changed your life quite a bit in real life yeah some of the things that have so explain 1:54:16 to the the folks here because you and I already chatted it just happened backstage I just had to fight these people off me to get out here no but 1:54:23 when people meet me they really freak out because it's like we have a relationship I'm saying their names I'm saying my name and then I'm giving them 1:54:29 their music but I'm giving you context around what you're listening to we're doing storytelling around what's 1:54:35 happening I'm a part of their lives because it's something that's on demand unlike other celebrity you may know them 1:54:40 by seeing them on stage or the court or whatever it is they do the difference here is they feel like I know them and 1:54:47 that's the biggest difference well in the year and a half that X has been out in the AI form how has your perspective 1:54:53 on your own identity changed honestly it's about the human connection and how 1:54:59 powerful that is of a force in this in this world that's what people are connecting with ultimately AI is the 1:55:05 conduit to people to connect to the human version of me this is a digitized version of me and my personality does 1:55:10 this feel like immortality in a certain sense I'm going to live forever it's I your voice right live forever me I think 1:55:17 that's I think do you but do you conceive of that like I did I thought about that before it came came out I 1:55:22 said this is this is a legacy project the truth is it's going to outlive me and that's that's the reality of which I 1:55:28 think is awesome no I do too but I'm just thinking like what extra thought then do you put into the creation of the content as a result of you understanding 1:55:35 is going to live beyond your life I take it really seriously I take the craft of Hosting very seriously I take the craft 1:55:41 of real authentic connection with people very seriously so I lean very heavily 1:55:47 into that knowing how big of a deal it is for people that I'm with them throughout their daily lives like that so I I thought a lot about that what I 1:55:54 would say the types of things so we so I was able to help shape the product and what's out in the world now you're at 1:56:00 our highest moments in life and you're at our lowest moments that's what that's what music is it soundtracks your life 1:56:05 crying into the phone all of it I'm there for you oh I know I love that this is the human connection we're talking about Xavier it's awesome to have you 1:56:12 here with us we're going to thr a playlist our axios events playlist scan the QR code enjoy the break we'll see 1:56:18 let me tell them how to scan it open your Spotify go to search click on the camera then it'll take you to the 1:56:24 playlist and if you want to check me out if you don't already check me out shame on you Shame Shame check it out search 1:56:31 AI DJ or DJ in Spotify and I'll be there I'll be honored to be with you each and every day as your personal AI DJ Make 1:56:38 Some Noise New York City thank you thank you hope thank you all right guys enjoy the break we'll see you in a little bit 1:56:45 see you [Music] 1:57:03 [Music] 1:57:25 [Music] 1:57:43 [Music] 1:58:08 [Music] 1:58:27 [Music] 1:58:33 [Music] 1:58:39 [Music] 1:58:52 [Music] 1:59:14 [Music] 1:59:39 [Music] 1:59:45 [Music] 1:59:58 [Music] 2:00:20 [Music] 2:00:45 [Music] 2:00:51 [Music] 2:01:04 [Music] 2:01:26 [Music] 2:01:51 [Music] 2:01:57 [Music] 2:02:10 [Music] 2:02:32 [Music] 2:02:57 [Music] 2:03:03 [Music] 2:03:16 [Music] 2:03:38 [Music] 2:04:03 [Music] 2:04:09 [Music] 2:04:22 [Music] 2:04:44 [Music] 2:05:09 [Music] 2:05:15 [Music] 2:05:28 [Music] 2:05:50 [Music] 2:06:15 [Music] 2:06:34 [Music] 2:06:56 [Music] 2:07:21 [Music] 2:07:26 [Music] 2:07:40 [Music] 2:08:02 [Music] 2:08:26 [Music] 2:08:33 [Music] 2:08:46 [Music] 2:09:08 [Music] 2:09:32 [Music] 2:09:38 [Music] 2:09:52 [Music] 2:10:14 [Music] 2:10:38 [Music] 2:10:44 [Music] 2:10:58 [Music] 2:11:19 [Music] 2:11:44 [Music] 2:11:50 [Music] 2:12:03 [Music] 2:13:14 please welcome our executive editor of axios live and new platforms Sarah K helani goo 2:13:23 hello hello everyone I hope you enjoyed a short break please take your seat 2:13:29 we're excited to kick off the second half of our program and I'm going to try to channel 2:13:35 that Spotify DJ energy I'm not sure I can quite Do It um but please come on 2:13:41 over I want to make sure you all can hear the next part of this program and I know the seats were hard 2:13:47 to come by in the first half um I'm Sarah K helani goo I'm the 2:13:53 executive editor of axos live axos live is designed to take our expertise our 2:13:59 journalism and bring it to life to the stage and to let you become part more 2:14:06 connected to the news shaping our world and why it matters now if I could take a quick uh poll I'd love to know who has 2:14:14 this is their first time in an axios event awesome awesome well I'm excited 2:14:20 to share with you today I hope you're having a great time I want to share with you about how we're growing axios live 2:14:27 our events around the country and around the world and if you're enjoying what you're experiencing today let me share just a 2:14:34 moment about where you can expect to see axios next today of course is focused on 2:14:39 AI and how it's shaping our world but of course technology is not the only topic we cover at 2:14:45 aios so our Newsroom is focused and our events team have no shortage of topics 2:14:50 that we're cut covering and just this summer you might want to take a quick photo of this this is where we'll be 2:14:55 just in the next few weeks and months in a few weeks we're going to France um we're going to can to talk about sports 2:15:02 media business we're also going to have be of course at the political conventions both the RNC and the 2:15:09 DNC um and we're also going to the Summer Olympics for the first time this year which will be really exciting we're 2:15:15 really looking forward to that and the fall stay tuned Ena told you about this we're also coming back to AI but this 2:15:21 time in San Francisco at the end of the year and in between we're going to be um also launching uh having a going back to 2:15:27 New York for our uh United Nations General Assembly and climate week lots of different topics to explore and if 2:15:34 you're interested please sign up to hear about our next upcoming events at 2:15:39 axios.com events now one more quick announcement don't forget if you haven't already to 2:15:47 you have a a QR code on your badge scan it and get to check out and play with 2:15:52 our meta AI imagine tool if you take a make a cool AI visual with it and send 2:15:58 an email to us we're going to pick three winners who at the end of the program today are going to win pair of Rayband 2:16:05 meta smart glasses and I hear there might be another few giveaways in 2:16:11 addition to that so stay tuned so without further Ado let's get on with the show 2:16:19 [Music] 2:16:27 and next it's me axio one big thing host Nyla [Music] 2:16:34 budoo how are [Music] 2:16:39 you hi everyone I'm Nyla budoo as you just heard I'm the host and editor of the axio one big thing podcast where I 2:16:46 talk to leaders shaping conversations in business politics and C culture and I'm 2:16:51 so pleased to let you all know that this next conversation will be broadcast on this week's podcast so thank you for 2:16:58 being part of not just being here for the AI Summit but being part of our live audience for this next segment and by 2:17:05 the way you can subscribe wherever you get your podcast one big thing this episode will be out tomorrow now to our 2:17:11 guest for more than 35 years the Robin Hood Foundation has provided millions of 2:17:16 dollars of Grants to fight poverty in New York now the organization famous for using investment principles to fight 2:17:23 poverty is enlisting AI in the fight please join me in welcoming the CEO of the Robin Hood Foundation 2:17:30 [Applause] [Music] 2:17:35 richb how are you how you doing good all right thank 2:17:41 you for being here thank you for having me so let's start with the big picture 2:17:46 when we're thinking about the US e our economy on paper looks really good right 2:17:52 now if you think about unemployment rates inflation easing somewhat from 2:17:57 your perspective how does that look for people who are at the bottom of the economic ladder so first of all thank 2:18:03 you for having me it's great to be here uh as you said if you look at many of the indicators of the economy the 2:18:09 economy is coming back strong uh but unfortunately that progress Mass the 2:18:15 Deep inequities that underly our economy which were existed before the pandemic and which persist during this recovery 2:18:22 so you know at Robin as you mentioned you know our mission is to fight poverty in New York City we want to create lifelines to Economic Opportunity for 2:18:29 those that have been left out of the economic mainstream here in New York City uh and what we see in New York City 2:18:34 is that indeed over the past several years poverty has become both larger and 2:18:40 more intense uh we do a survey every year called the poverty tracker which tracks the experience of low-income New 2:18:46 Yorkers and what we saw in the most recent year that the number of people living in poverty increased by 2:18:53 500,000 from 1.5 million to 2 million New Yorkers living in poverty 25% of all 2:18:59 children in New York City live in poverty uh and indeed even that number understates the problem because if you 2:19:06 uh even if you earn double that amount if you live 200% of the poverty line in New York what is that you are still 2:19:13 $888,000 for a family of four using something called the supplemental poverty measure which is a more robust 2:19:18 measure of of poverty uh if you earn $888,000 for a family of four in New 2:19:23 York you are twice as likely to struggle paying the bills to have to make a choice between paying rent or seeking 2:19:30 health care or buying food uh and that is the majority of New Yorkers so um in 2:19:37 some ways the city looks strong but that strength hides deep inequity you talk about unemployment yeah the unemployment 2:19:43 rate in New York has uh really rebounded but for black New Yorkers a poverty rate is about three times the poverty rate of 2:19:50 white New Yorkers so we continue to be in a moment where those who are at the margins of the economy are continuing to 2:19:56 struggle and so you all have opened an AI poverty challenge that will provide three1 million grants what's the hope 2:20:04 with that so uh for us when we think about what does it take to move people 2:20:10 out of poverty um the things that make the difference between opportunity and 2:20:15 not being have access to opportunity or things like education like you have access to a great education 2:20:21 do you have access to high paying jobs that are going to allow you to sustain your family do you have access to basic 2:20:27 financial support so that you have housing and food and the things that provide stability so that you can build 2:20:34 a better future for yourself do you have access to Capital uh that you need to buy a home or to invest in your future 2:20:41 all of these things are part of the economy that are deeply disrupted and impacted by artificial intelligence so 2:20:48 as an organization whose mission is to fight poverty we have to be an organization whose mission is to think 2:20:54 about how do these emerging Technologies how are they going to impact those living in poverty and uh one of the 2:21:01 things that as we as an organization have tried to understand more about how these things intersect uh all the you 2:21:08 know every time you hear about AI you sort of hear about this very bipolar view of the world either everything's 2:21:14 going to be Eden and Nirvana or everything's going to be uh Skynet and Devastation and I think ultimately uh 2:21:21 like any other thing that happen in our world um whether AI will be a force for 2:21:27 good whether it's a force that unleashes opportunity or whether it's a force that oif in inequity is going to be 2:21:34 determined by what people do um what are the choices we make and so for us this AI opportunity challenge is one way to 2:21:41 put the call out to all the incredible people out in the world including the technologists that are building the 2:21:46 future to ask them what would it take to build tools that ensure that AI is truly 2:21:52 a tool for opportunity and not only a tool to retrench inequality so actually 2:21:58 next week on the podcast my guest is rashma saani the founder of girls who code and moms first and we're actually 2:22:04 going to be talking about how much the narrative around AI is causing fear or hindering people in the same way that 2:22:11 she saw with young women and people of color uh who originally left out of careers in Hightech from you where you 2:22:18 sit do you think this is actually happening with AI yeah absolutely I I I you know I think part of what happens is 2:22:24 when you lead with a narrative that says either everything is going to be great or everything's going to be horrible 2:22:31 what that does is it removes any need for agency it means that um why engage 2:22:37 when it doesn't matter what's going to happen and so I think part of the message is to reinforce that it does 2:22:42 matter what happens it matters what we do we can sort of decide the future that we're building together um and the part 2:22:49 of the challenge in the what's happening with the economy right now is that uh the people who are most vulnerable uh 2:22:57 are just disconnected from these decisions about what's going to happen with our economy they're disconnected from what's happening with AI uh you 2:23:03 know we in New York City a city which is made incredible strides in terms of internet access in a place like the 2:23:09 South Bronx 40% of families do not have access to highspeed Internet uh we 2:23:15 everybody's probably seing the news about the American connectivity program a pandemic era policy which is designed 2:23:21 to connect people uh to high speeed internet around the country uh tens of millions of Americans are at risk of 2:23:26 losing highspeed internet as that um as that money uh disappears so what is 2:23:33 underlying all of this is that people who uh stand to win or 2:23:40 lose um the voices of the low-income populations are just not part of this conversation and so one of the thing 2:23:46 that really drives our thinking is how do we Center those living in poverty how 2:23:52 do they become a part of building the future how do we make sure that um we're building tools that are actually 2:23:57 responsive to the needs of people living in poverty so that um AI can be a force 2:24:03 for good and not a force to further inequality so how are you already seeing this happen like what are examples where 2:24:09 you're seeing this when it comes to your work yeah we're we're really um fortunate that we're able to we're 2:24:15 investing in some really amazing organizations and programs and and I do think that theme of many of them is how 2:24:21 do we connect people um to decision-making how do we get these tools in the hands of people so that 2:24:27 people have an opportunity to build what they need so one partnership that we just announced yesterday that we're really excited about is a partnership 2:24:33 with Teek NYC Foundation um which is really trying to address the problem of 2:24:39 nonprofit organizations in New York all of whom here every day you got to be using AI you got to be using AI but what 2:24:45 does that mean how do I use AI um and how do they start how do you start how you understand the use cases how do you 2:24:52 understand how these emerging Technologies can actually make a difference for the way that you work or the people that you serve so we're 2:24:58 really excited about this partnership and the idea is to activate um Tech nyc's network of 700 plus Tech 2:25:05 Industries in New York and to connect them with organizations schools nonprofits that are on the front lines 2:25:11 of meeting the needs of those living in poverty to help them understand uh how what are the use cases how can you 2:25:17 improve your service delivery um and another example that I'm really excited about is playlab AI this is a national 2:25:24 nonprofit that again is trying to connect uh people who work with children 2:25:31 um to these AI tools and basically what play playlab AI does is it um creates a space where teachers can come together 2:25:38 in learning communities and play play with uh chat GPT play with Claude play 2:25:44 with these tools and think about how do we solve how do we build tools that solve the need of my students in my 2:25:51 classrooms things like uh building a chat bot to help uh parents uh whose 2:25:57 kids are after from school to engage them to address chronic absente them tools that are helping to provide 2:26:03 formative Assessments in real time for essay writing in a way um that leverages the skills that the teacher wants to 2:26:10 give but can give that feedback much more quickly than the teacher would have to do if they had to read UH 60 essays 2:26:16 but but again uh not having corporate executive deci what people need but 2:26:22 actually having the people that we want to use these tools decide what they need uh one of the thing that we do at Robin 2:26:29 Hood uh is a a team called blidge lab it's essentially an incubator for technology enabled Solutions uh to 2:26:36 poverty and uh Central to our work at bluid Labs is something called the design Insight Group which is a group of 2:26:44 New Yorkers with lived experience of poverty uh who provide feedback uh to 2:26:50 the Tech Founders about the tools that they're creating to help solve their needs they help set priorities they 2:26:56 provide feedback to Founders they do early product testing you know one of the thing that we always used to hear at 2:27:02 bluid labs when people wanted to come and sort of create Financial tools for people living in poverty they would um 2:27:07 say well I'm going to help you create a budgeting tool we got a lot of uh information about budgeting tools but when you talk to people in the design 2:27:13 inside group a dig they'll tell you we don't need budgeting tools I I'm a ninja with the budget I can make a dollar 2:27:20 stretch more than you could ever tell me how to make a dollar stretch what I need help with is how to understand what my 2:27:27 um SNAP benefit balance looks like and how do I actually maximize those benefits what foods are eligible for 2:27:34 Wick um one of the program that we're funding through Brew's lab that I'm really excited about is something called 2:27:39 climate. us which is essentially it's almost like a a cpilot for social 2:27:45 workers right like so the idea is if um if you are eligible for medical benefits 2:27:51 uh for dis medical disability benefits it's really complicated uh we know that nationally any given year uh 3.5 million 2:27:59 Americans who are eligible for those benefits don't get it at a loss of about $3.5 billion a year um sorry $300 2:28:07 billion a year in New York alone we're talking about $3.5 billion in loss benefits but what clayt is trying to do 2:28:13 is to pull up people's medical records evaluate what benefits they're eligible for uh turn that into into an 2:28:20 application and to really improve the process of applying for and appealing 2:28:25 denial of benefits uh but these are all real world solution that are driven by actual people's needs uh you know one of 2:28:33 the um it's almost a cliche in our world to speak about proximity that people who are closest to the problem are the best 2:28:39 position to solve it but I think you're probably hearing such Amazing Stories right like when you're saying these 2:28:44 people like what to your mind like what stands out like I think that's a very Vivid example of the snap benefits like 2:28:51 when you think back to these conversations what stands out from what people have said to these Tech Founders about what they need well part of it is 2:28:57 is the the the excitement and the joy of people thinking about how to center uh 2:29:04 how to center their challenges um and I think we're all so used to living in a world where things are done to us and I 2:29:11 think what we hear from dig members is the joy of being empowered of being asked what is your life really like what 2:29:18 do you need what would it take to be able to move your family from a place of vulnerability to a place of stability 2:29:25 ultimately to a a place of wealth building and opportunity um and so it's really everybody wins of course uh the 2:29:32 founders win and sometimes either for-profit sometimes they're nonprofits but they win because they're building products that people actually need um 2:29:39 but there's a tremendous sense of empowerment to the people in our dig who understand that they're being treated like experts in the realm of expertise 2:29:47 uh and I think the whole city wins the whole world wins and we can have more development happen that way so to your 2:29:53 mind you see ai's best use case for your work as empowering it's incredibly 2:30:00 empowering I mean so I think one of the things that you know as I said before people either have a very optimistic 2:30:06 view of this or very pessimistic view simpl it's way too simplistic and I'll admit I'm an optimist by nature and I 2:30:11 think there's good reason to be an optimist when it comes to the power of technology to transform lives we've seen 2:30:17 over the last 200 years the percentage of America of people in the world who live in abject poverty have gone from 2:30:23 80% to 10% a lot of that is driven by technological innovations and Agriculture and housing and food so it 2:30:30 is good to be optimistic um but that doesn't just happen it will happen if 2:30:36 people make the choice to Center the needs of all uh Americans all people in 2:30:41 the world uh when they're deciding how to build and Implement these tools and so I'm optimistic but I'm optimistic in 2:30:48 part because I get to spend my time working with these incredible leaders and entrepreneurs who are really trying 2:30:54 to solve The World's problems uh and so if we can Empower those people and empower the people that they're trying 2:30:59 to serve uh I'm very bullish on the future uh exis we like to end on one fun 2:31:05 thing so rich you just had your foundation's annual fundraising Gayla in May it was Matrix themed how did you 2:31:12 guys you guys went all out for this how did you come up with that oh well we we always go all out but you know we um it 2:31:17 really well it came out of our work with AI you know in February we had a summit on AI and poverty um and uh again we 2:31:25 talked about the negative it made it think about well the Matrix can you hard to think of a a movie more about the uh 2:31:30 the harms of AI renom Muk uh and so you know we ask the crowd at our benefit to 2:31:36 the blue pill to take the blue pill um and to take the pathway to solving problems and we raised $58.5 million for 2:31:43 New Yorkers in need that night uh it's a lot thank 2:31:51 it's a lot of money uh but it only scratches the surface of need um and so we need more than philanthropy we need 2:31:58 businesses we need Civil Society we need government to all come together and to put Americans uh put low-income New 2:32:05 Yorkers put people across the world who are suffering first if we do that again there's no limit to what we can do well 2:32:10 I'm afraid we're out of time but I think that's a very good place to end the CEO of the Robin Hood Foundation Rich Fury 2:32:15 thank you so much much good to talk to you thank you 2:32:21 [Music] 2:32:49 [Music] 2:33:04 welcoming back Ena [Music] 2:33:11 freed are the shoulder pads supposed to go under the check you may have noticed I added an 2:33:17 accessory it's not a fashion statement well it's not just a fashion statement um it's 2:33:25 actually part of an AI project between AWS Amazon web services and the NFL it's 2:33:31 part of a broader campaign they've T the digital athlete and it's part of an attempt to mitigate against injuries and 2:33:37 more here's a quick video to explain more and I will be back in just a 2:33:43 second in the NFL the game is constantly evolving and S our ways to better protect players powered by AWS the NFL 2:33:51 created the digital athlete an injury prediction tool by analyzing millions of data points generated by a player's 2:33:57 every move AWS aai and machine learning are helping to identify when players may be at an increased risk of injury 2:34:03 enabling teams to develop individualized training and Recovery regimens the NFL is focused on the 2:34:09 health and safety of every player it's a key to building a better game 2:34:17 okay okay so that's a little bit of glimpse of what the NFL is working on here to chat with me more about how this 2:34:23 works is Jeff Miller an executive VP for player safety comms and public affairs at the NFL and Julie Souza the global 2:34:30 head of sport at AWS Jeff and [Music] [Applause] 2:34:37 [Music] [Applause] Julie so we all saw the the pads and 2:34:42 that's great that's one place that the NFL is collecting data but that's that's only a small part of it where all are 2:34:48 you collecting data and what are you using it for so much data um in addition 2:34:54 to the pads every player so 2,000 players a week wearing GPS locators you see it when you watch the games you see 2:35:00 how fast players are running or the routes that they're running and when they're open or not or what the quarterback sees that comes from that 2:35:06 but in addition every piece of equipment they wear specifically helmets but cleats and shoulder pads too as well as 2:35:11 really importantly the vast majority of cameras that we that you see angles from we're getting data from including even 2:35:19 last year when you saw um cameras in the pylons at the goal line all of this is 2:35:24 contributing to the work that we're doing in collecting and aggregating all of this information uh to help us make 2:35:29 the game better we're going to talk about what you do with that but before you can do anything with it that's just a ton of raw data and before you can use 2:35:37 that much data you have to make something out of it and I assume that's where you and the AWS team come in Julie 2:35:42 yeah I mean and just to underscore the point about how much data um and Jeff alluded to all the different sources of 2:35:49 it but from the Optical tracking 38 cameras in in the venue that's 6.8 million frames of data per week that 2:35:55 we're getting um and I think we have a video that shows kind of what's coming in and how you make it but when it comes 2:36:03 to that much data like how do you even know what to look for how do you assess the patterns what do you do with that 2:36:08 Amazon just so you know AWS I mean this is your job as head of sport you're doing this for a bunch of different 2:36:14 sports and obviously each sport has different needs with the NFL there's a lot of safety issues 2:36:20 how do you start sense of that data and of it I mean it's always working backwards from the questions we're 2:36:26 trying to solve right so and Jeff we can talk about this a little bit later too I mean concussions are a real concern in 2:36:31 the NFL so how do we first detect them right and so that's where a lot of these 2:36:37 different tracking sources come into play and then how do we deploy technology like computer vision and AI 2:36:43 nml to help look for those patterns in the data to help us isolate and then create models to predict and protect 2:36:52 right so risk mitigation modeling is a big part of the work we're doing uh for digital athlete with the NFL and I think 2:36:59 we have a video if we can play that um this shows some of that data coming in 2:37:05 um so you're measuring stuff that's useful for safety you're measuring stuff also I think that's useful for fans 2:37:11 that's useful for you know gambling and betting is an increasing Revenue source for all the sports leagues um a lot of 2:37:19 this data serves multiple purposes right yeah I mean that's what's exciting about it it sort of started and one of the 2:37:24 inputs is the nextg stats data that helps feed sort of on if you're watching 2:37:30 it on on Thursday Night Football on on Prime video the prime Vision cast right all of those analytics and defensive 2:37:35 alerts and things like that so there's absolute broad application player health and safety rule development rule changes 2:37:43 I mean you can talk about this stuff right yeah because one of the things that's critical is not just knowing 2:37:48 there's a lot of head contact and concussions but what do you do about it and so talk some of the ways that you 2:37:54 are taking action first excellent job on Cross promotion for Thursday Night Football so that's appreciated always uh 2:38:01 but but also what you saw there for a second is exactly what our clubs take a look at what our trainers look at our doctors look at literally on a daily 2:38:07 basis they have an interface looking at every player all their loads what they've done in practices what they've 2:38:12 done in games what is their injury history what equipment are they wearing and from that with aws's help we can 2:38:17 predict their risk profiles what point is somebody running hot which receiver do you need to pull back on that week 2:38:23 who's been exposed to more than they should and we can start to govern how we manage them now in addition to that I 2:38:29 think we have another video let's take a quick look at that because it shows what you're talking about I think so this is 2:38:34 a video from a game um and what you're seeing in the red if I'm not mistaken is 2:38:39 head-to-head contact that's our head detection system so we used to literally go through manually not me but somebody 2:38:46 would go through and count every head contact as we were looking at ways to better understand the magnitude of head 2:38:51 contact and the likelihood of injury specifically concussions share that information with helmet manufacturers so 2:38:57 they could design better helmets against it instead we taught the computers to do it so literally every time somebody hits 2:39:02 their head in an NFL game we know we share that back with the clubs with the coaches we share it with the equipment managers we share it with the equipment 2:39:09 manufacturers to say build us something better than what we've had before given the fact that different players 2:39:14 experience the game differently suffer injuries at different rates and in different ways and it wasn't that long enough ago that everyone wore a similar 2:39:21 helmet then they have position-based helmets now they have them tailored to each individual then as you talked about 2:39:27 another level that it's being used is so that teams know at the individual level 2:39:33 how each of their players is kind of how much what their exposures are and what 2:39:38 the risk is of them suffering an injury now it's not a clinical tool it's not a diagnostic maybe someday maybe but we 2:39:44 know the velocity of those impacts the magnitudes the closing speeds we know the direction the impact how frequent 2:39:49 all of the things you might imagine and to learn that a quarterback suffers a concussion very differently from an 2:39:55 offensive lineman but until this past year they wore the same equipment made no sense so now we have eight new 2:40:00 position specific helmets for the lineman for the quarterbacks next year for hopefully folks in the secondary 2:40:06 because they experienc things differently but not only that getting back to the rules part that Julie mentioned that same platform that you 2:40:12 saw it was the foundation for what will be the biggest change in the NFL in many years which is the change to the kickoff 2:40:17 we were able to identify the kickoff as the highest rate of injury and the highest concussion rate of any play not 2:40:23 the highest number but the highest rate and so when you took a look at that and then tried to design a play where you 2:40:28 would minimize those areas where there's the highest risk exposure led to the pretty different changes you're going to 2:40:34 see uh in a couple months so when kickoff happens in September kickoffs won't look quite the same and part of 2:40:39 the reason is what you were able to get from the data we saw where the biggest risks are to the most players in terms 2:40:46 of injuries specifically concussions the spaces in the spe in the areas on the field where they hit one another specifically the closing velocities how 2:40:53 do you extract those risk areas and yet have a play that is still relevant and 2:40:58 maybe even exciting so with all that extracted we knew what the issue areas were and could start to address those 2:41:03 and figure it out the fans are going to see it in August and September I am sure that they're going to react hopefully 2:41:09 well I think it's going to be really exciting but it was all driven by some of the data that we're helping um that 2:41:14 AWS is helping us collect and and and um work with most of what we've seen are different ways you know both at the 2:41:20 rules level Team level individual level to make the game safer I'm curious do 2:41:25 you think this gets us as far as we need I love football I love watching football and it's concerning to me that it seems 2:41:32 to have long-term risks for all the players that are involved do you think technology will get us to a point where 2:41:38 we can say football is safe for our sons and daughters to play well it's what you do with the technology that gets you 2:41:44 there I mean as I stand here today compared to 3 years ago concussions are down 25% across the NFL that's great 2:41:51 hamstring strains are down 40% across the NFL because we redesigned the beginning of training camp as a result 2:41:57 of the data that we were collecting because of the sensors on all of the players so can you get to the fact never going to be any injuries of course not 2:42:03 there's no Finish Line here but the equipment will improve the training will improve the gameplay will improve 2:42:08 witness the rules and we're going to end up with a better and safer game for the players one last thing I'll say on to that is that it's not limited to the NFL 2:42:15 I think this has broad application not only to other sports but other Industries like military and training 2:42:20 and things like that so I think a lot of the the forward work that the NFL is doing will have broad application um 2:42:26 Beyond just game of football all right so you heard it here first be sure to change your fantasy team accordingly 2:42:33 thank you so much to Jeff and Julie and I did want to just put one more reminder if you did want to take part in the AI 2:42:39 photo contest to win The Meta smart glasses you do have to not only create the image but email it to us if you're 2:42:45 having any trouble come to the registration desk someone from ax will help you do it we've gotten a bunch of 2:42:50 really cool entries but we'd love to see more um so again you just use the QR code on the back of your badge and uh I 2:42:57 will be back in a minute with our last couple of interviews and we've got a little bit more for you so thank you 2:43:03 [Music] 2:43:28 hey hey hey hey hey hey hey 2:43:36 [Music] 2:43:53 introducing to the stage again hope 2:43:59 King I'm back and then props are there'll bring all right so we've been 2:44:04 talking a lot about themes around Sports we just heard we talked about music um 2:44:10 we have not talked about the money yet where is the money going who has the 2:44:15 money and show exactly show me the money everybody right um this next panel will 2:44:22 talk about where the opportunities are right here in New York uh for those of you who are in the East Coast West Coast 2:44:28 battle like I admittedly I think West Coast has better food and weather um but we have Bodega cats and coffee and great 2:44:35 pizza and we do have a really robust AI ecosystem which we're going to talk about in this next segment with my next 2:44:40 two guests first is Grace isord she's a partner at Lux capital and one of her 2:44:45 portfolio companies osmo is here the CEO and founder Alex witko uh please welcome 2:44:51 to the [Music] 2:44:57 stage all right Alex you have a couple of things 2:45:04 that you want to show us before before we get into our conversation and the first of which is just describe what osmo is it's a company designed to 2:45:10 digitize smell and you're trying to develop next gen Aroma molecules which I 2:45:17 didn't know I needed a next gen AR molecule um so let's go through the premise and sort of where you're going 2:45:24 with the ultimate products that you hope to develop absolutely so let me see if I can show you a little bit of a visual 2:45:29 aid there so we're in an era where we've digitized 2:45:34 several senses and a lot of what we've heard about today is built on top of that we've forgotten that right so 2:45:40 vision and hearing have already been digitized but not Evolution's first invention smell that is our first and 2:45:47 most primary most emotional sense and computers have no idea how to do this so 2:45:52 how how would we do this right I I think I think it's important to look at how other senses have been digitized it's 2:45:59 really always in three steps you got to read the world turn the real physical world into digital signals so with a a 2:46:05 lens and a sensor and then you have to map it like what RGB is for color that lets you organize the information and 2:46:12 then you need to write it back out again like with a printer or with a display like you're looking at and this is how it's worked for vision and there's the 2:46:18 same Arc for hearing right you've got a microphone to read you've got mp3s and the 4A bases to map and then you've got 2:46:25 speakers to write it back out again what we're doing at osmo is we're building the technology stack to read map and 2:46:32 write smell and we're building this from scratch and we started with a map uh 2:46:37 we're working on read and uh I'll actually have a chance to share what we're doing with WR in just a second but 2:46:44 um you know if you can do this the question is um you know well first of 2:46:49 all we've made a lot of progress I won't go into this in too much detail but you can come find me afterwards but we're 2:46:54 we've got fantastic a fantastic team got fantastic science fantastic investors uh 2:47:00 and we're about a year and a half old right now and we're well into actually being able to digitize the sense and we'll have more to share uh as the year 2:47:07 goes by but you what would you do with it I want to leave you with you know some of the different uh go to market 2:47:12 areas and different areas in which society could be improved and uh you know obviously there's consumer Brands 2:47:18 right there's the beauty side and then there's the truth brand right so uh can we detect things that are harmful so one 2:47:24 thing that we've actually been doing is working with the bill and M The Gates Foundation to use the same AI models that design new fragrance molecules and 2:47:31 we should actually smell them in a second to find molecules that don't smell necessarily great to humans but 2:47:36 smell really bad to mosquitoes so we've developed a new generation of insect repellents and with the bill and M Gates 2:47:42 Foundation we hope to save lives with that right so not only can we build a strong business on the capabilities that 2:47:47 we have today but we think that we can do some incredible good in the world and there's obviously industrial applications and government applications 2:47:53 as well as uh entertainment in the very very long run so that's uh a bit about us a very very high level but um 2:48:01 computers will have a sense of smell this will happen I think osmo is the company that will make this happen much 2:48:06 faster than it would have if we had let it happen organically and people were going to live H happier and healthier lives as a result of that so just a bit 2:48:13 about osmo but do you want to smell some of what we invented this is not like smell vision this is real this is real 2:48:23 stuff all right so a little bit of a like a perfume sampling lesson that 2:48:29 we're going to do here so these are called blots and you dip them into um fragrance samples in order to smell them 2:48:35 so what I'm going to do is open my magic little box of AI generated smells here and uh there's no magic here these are 2:48:41 each just vials um now what's in here are molecules that were invented by AI 2:48:50 so we taught a machine Learning System and AI system to predict what a molecule will smell like digitally and we 2:48:57 validated this and now the final validation the you know the proof in the pudding is in the smelling so I've got a 2:49:04 lot to choose from but I let me let me show you a collaboration between traditional perfumery so we have a 2:49:10 master perfumer he's done work for Tom Ford and abber groman fish and AI right so what it looks like when Ai and human 2:49:16 collaborate would that you want to smell that okay great so let me get two blots here actually three for me as well and 2:49:22 I'm going to give you uh a hybrid AI generated and human craft invention and 2:49:30 I want you to just tell me what you smell do you like it um what do you perceive in it so take one and pass it 2:49:37 on thank you I'm spelling Rose Floral sense it 2:49:42 floral yeah my allergies are acting I'm sorry clear your nasal passageways I wouldn't know this is AI at for what 2:49:49 it's it's very natural right so like smells like for folks you know like Joe Malone perfume the rose white like the 2:49:54 English pair yeah so it's like a white floral very sophisticated and like it AI generated and synthetic doesn't have to 2:50:01 be unnatural in fact I think this smells very beautiful and very natural and you can come find me and we can smell this 2:50:07 together if you want Pro tip when you're smelling these things bend it a little bit and then you can place it down so 2:50:13 that it doesn't uh contaminate the surface so um let's do one more um 2:50:19 and this is kind of a different odor family also one that I think is pretty beautiful um let me pull out this one 2:50:33 y I'm sorry we don't have samples for I really I asked for like popcorn movie 2:50:38 theater popcorn right so that it would permeate but we're not quite there yet thanks we'll put it in a fan next time I 2:50:43 mean in the limit this is going to be something that you can experience at home what do you think hope what's the sense I'm I'm I'm like my allergies are 2:50:50 really bad I picked her wrong day yeah I know I I like I took an algo yesterday and I thought it but I'm sorry um no I 2:50:57 this this for sure smelled really beautiful like a garden and I'm having a harder time with this one CU to me 2:51:02 there's like a uh like a smokey wood yeah kind of a musk Andy dewy yeah and 2:51:09 there's a bit of vanilla in it as well so one of the you can't actually make this scent without inventing new 2:51:15 molecules so some of these smells exist in nature but perfumery hasn't actually been able 2:51:21 to recreate it it's like painters haven't been able to paint this picture without the new pigments all Factory 2:51:26 pigments that we've invented well there's a sustainability element to this is that unfortunately some of the ingredients that the you know scent 2:51:34 masters of the world are using they are dying uh and so in order to have those 2:51:40 scents live on you have to map them now so that we can you know make sure that they're in the fragrances that we have 2:51:47 um so that's just you know something that I came across that I felt was pretty compelling and you know a sad reason for why we need this um but I 2:51:53 think you know ultimately I I made a joke about smello Vision right I mean this is like the concept has sort of been around um but ultimately you know I 2:52:01 I I hear three different types of of use cases and just Grace just from your perspective is an osmo sort of 2:52:10 application of AI something that's very typical that you're seeing now is it in the majority or is it really a a pretty 2:52:16 you know rare case that you see or a use case of of AI well Lux is a deep Tech VC so we do look for things that sound 2:52:22 crazy and some of the craziest ideas in fact are some of our best investments this actually came from a thesis we were 2:52:27 searching for to recreate the sense of smell you know you can recreate taste or motion uh but and really haven't seen 2:52:34 that happen yet uh so we're looking for the intersection of AI with embodied Sciences so that could be you know 2:52:40 creating a sense of smell that also could be you know automation of Robotics or Ai and the life sciences so biology 2:52:47 chemistry Etc well in the you know nearly 2 years since chat gpt's launch how have your everyday lives changed and 2:52:53 also your technology has it actually enabled any type of acceleration uh with 2:52:59 development of what you want to work on but Grace I want to just hang on with you for a second first I use it frequently particularly for synthesis of 2:53:06 research papers and you know Cutting Edge new Innovations um I've created like small little workflows to automate 2:53:11 daily tasks of synthesis of notes and set a reminder to come afterward uh but majority of it has been used from a 2:53:16 research perspective today have the pitches changed at all that you've seen a lot of pitches are saying the same 2:53:22 thing of we can use AI for this so then how do you sift through that right as an investor we always look for things that are n of one or technically unique right 2:53:28 so what can they do that no one else in the world can do that could be cutting edge research coming out of a research lab where I don't think there's actually 2:53:34 anyone better than Alex to create this company in the space and also look for that 10x differentiator right whether 2:53:40 it's knowing the worklow very deeply whether it's having the researcher who led the open source project that is at 2:53:45 the Helm of the company and Alex what about you has the recent developments impacted the way that you work on the 2:53:51 back end so I I was very fortunate to have worked at Google brain where Transformers were born I was there for 2:53:57 about six years where I started where the research originally came from which we spun out with Lux and with GV and um 2:54:03 I got to see that Arc begin very early and it was just clear from day one that everything was going to change or a lot 2:54:09 was going to change and so I've been slowly building this in and kind of having a little like assistant on the side there I need to learn about a lot 2:54:15 of new stuff every day and I can get a pretty good answer pretty quickly with chat GPT so it's it's kind of infused in 2:54:22 my life and my team uses it a ton so it helps accelerate code code reviews everything like it's kind of pervasive 2:54:29 in the company so on that code review point though has that helped you accelerate the development of your products does it help with mapping 2:54:35 quicker does it help with mapping more accurately would love to learn more about that so it's it's a sidekick for a 2:54:41 lot of workflow that we have in the company but what I find is that when you are at the edge of what's known or the 2:54:47 thing you're doing isn't in the data set you have to do it yourself and so we invest very very heavily in generating 2:54:53 massive amounts of scent data because nobody else has it full stop nobody else in the world has it and then we have to 2:54:59 build our own small Sharp Tools around that data in order to solve our problems 2:55:04 like design these molecules or many of the other experiences that that we're working on and uh I I find that there's 2:55:10 there's a balance right so if you want a general purpose helper um llms as they are are fantastic but when you start to 2:55:17 get into the edge of what's known or what's possible or your data actually is highly proprietary you don't always have 2:55:22 to reach for an llm or a Transformer and really the tool should be purpose fit for the job right a lot of cases people 2:55:29 feel recently ashamed that they used this to help them and obviously not the case it at your companies they feel like 2:55:35 it's that's cheating or they're taking some sort of shortcut I mean how do we dissuade people from thinking that way 2:55:41 well I'm a techno Optimist and I think if you're a venture capitalist you have to believe in the future um and so we actually see the opportunity right how 2:55:47 can this make your workflow better how can this lead to New Opportunities how can it save time in your workflow to 2:55:53 invest in other areas and think more broadly whether it's even as like a creative agent to ask questions and help you kind of push limits of your own 2:55:59 thinking all right so we're in New York you were doubling down in New York Lux Capital as you guys developed these molecules on First Avenue right uh and 2:56:07 29th Street I mean what is it about New York that differentiates this environment from the best coast SF all 2:56:15 North Cal I mean I know we're laughing but I I really I mean I'm from here but I I do love California well to start I 2:56:21 think Alex should share why he moved to New York because it ties in directly to to my thesis yeah we started in Boston 2:56:26 right that's I was there for 14 years it's the capital of a lot of chemistry I thought that's where we were going to build this company it turns out to do 2:56:34 what we needed to do there was very specific skills that only exist in New York and New Jersey right so to build oh 2:56:40 and New Jersey Shout Out New Jersey one the things uh so we need perfumers 2:56:45 there's actually more astronauts alive than there are perfumers so if you want to work with them you got to be here and 2:56:51 then there's very specific chemistry skill sets to make new molecules or to analyze those molecules and it's 2:56:56 different enough from what goes on in Boston which is all in like liquid you know blood and plasma but it's we need 2:57:03 to analyze the air and that is different enough and specific enough we had to be here actually specifically in that part 2:57:09 of Manhattan in order to get all the commuting distances correct and that's what we're seeing in our portfolio in 2:57:14 New York the talent and the demand is here right it's the home to 44 the 4 to 500 companies so people are coming here 2:57:20 regardless to talk to their customers to be working with their customers on top of that fantastic research Labs right 2:57:26 New Jersey again Princeton NLP Center is really good not too far away Colombia people forget that Yan Lon Chief 2:57:31 scientist of meta NYU silver lab based right here in New York Cornell Tech Sasha rush in our portfolio um a 2:57:37 majority of lux's AI Investments are based here in New York actually so hugging face the GitHub for ML Runway AI 2:57:43 for generative video P torch was born here P torch was born here um and so you're seeing also newer companies 2:57:48 whether it's osmo or modal some of our later Investments That more recently are moving here for the research for the 2:57:54 customers and for the talent yeah I mean look if I had this conversation in France they would have their own list of 2:57:59 people they could right but so I I bring that up also because I I do Wonder in your conversations you're having with 2:58:06 your own peers and as you're seeing the companies are there Geographic trends when it comes to talent and certain 2:58:12 parts of the world and when it comes to investor interest and also use cases what are some of those that you've been 2:58:18 observing over the past two years yeah I can start I actually did an analysis of the Lux AI portfolio and we saw a net 2:58:23 inflow and you have over 200 companies over 200 companies this was specific to our AI companies we have a company in Japan we have hugging face in Paris we 2:58:30 have several companies in the Bay Area we saw a net inflow of 57 of roughly 200 employees that had moved since 2020 to 2:58:36 New York and we asked them anecdotally you know why are you moving here in part it could be for personal reasons right I 2:58:42 just like living here more um it's not more scientific than that but other cases it's hey there's a really good 2:58:47 specific industry here that I want to talk to there's a really good researcher or research lab that I want to go deep in BIO and ml in particular is really 2:58:54 strong here the team that led you know met as AI uh practice in BIO is based right here in New York um and has more 2:59:00 recently spun out and so there's really really cool opportunities right on the Forefront Paris Tokyo things we're 2:59:06 seeing there Paris has a lot of good you know infrastructure Talent as well um and really strong AI Engineers out of 2:59:11 poly technique there's a hub going to be hubs in multiple cities right is going to be Hub New Y going to be a hub and so 2:59:17 the use cases will different slightly but it's really more on the companies that are bringing those communities together are investor mindsets different 2:59:22 depending on where they're located I think it's true um I think in New York the investor mindset is a little bit more practical a little bit more realist 2:59:29 right SF can tend to be a little bit more bigger picture dreamer I think it's great to have a bit of both um and I traveled SF frequently but I think both 2:59:36 bring different skill sets together um I think Paris is really looking for really top engineering and open source and M 2:59:41 and hugging base have been big pioneers of that I love that Alex uh you know for New York we're seeing a lot of folks who 2:59:47 simply want to be here right like young folks who are hungry because what we're doing doesn't have a name yet I mean 2:59:52 machinal faction is not really a thing we have to train people up from scratch and so bringing young excited hungry 2:59:59 talented people to New York has actually been pretty straightforward and pretty there been a 6X increase in just College 3:00:04 New grads coming here from 2019 to 2023 like one in seven I think of tech folks in the last four years have moved here 3:00:10 so there's a a trend you're seeing and now it's a matter of great companies like osma to continue to capture that so New York is going to be back I mean it 3:00:17 is back but it's going to be backer even back backer even backer um our last segment just really quickly on where AI 3:00:23 is going the frontier of it today and you know we don't have a huge amount of 3:00:28 time now to talk about it but how do you see it scaling I think we're really early I I think anyone is kind of fing 3:00:34 themselves to say they're seeing llms everywhere in widescale production today maybe in coding maybe in Customer Support is where I've seen it but we are 3:00:41 still so early from lm's working deep in production we're most excited is that intersection of the physical world so 3:00:47 that could be robotics and general robotic software automation that could be the Sciences as we have been discussing a bit today um and it could 3:00:52 be the New multimodal Frontiers but I think you're going to see AI integrated both in the physical workflows but also 3:00:58 in how they're manifested so audio text video all together and not in the chatbot format that a lot of our used to 3:01:04 and I love that today both my segments were about bringing AI into the real world you are clearly working on a form 3:01:11 of this where I could email sent is that the end goal or I DM sents to somebody 3:01:17 be like hey this is what I smell like right now is that the Future No I mean really is that the future or no it's 3:01:23 it's going to happen right you need to read mapping right but once you have all those things fused together new forms of 3:01:29 entertainment that will surprise us will be possible we need to build the Tex St that's step one yeah all right uh one 3:01:35 fun question that we're going to end with um so when open aai announces a partnership rolls out a new product or 3:01:42 the CEO tweets a pronoun which maybe happens to be the name of movie like what goes through your heads when when 3:01:48 that happens I'm not surprised um in fact open ey will have a product for everything and a model for everything 3:01:54 and so to me that's not the the question the question is more how good is that product how good are they actually applying that product in the workflows 3:02:00 of users and so great if you have a really cool voice model for example how easy it is it for someone to implement that into their product I'm going to 3:02:07 have to explain this to my mom in some shape or form right so I I'm thinking like okay what is she going to see in 3:02:12 the world and how like cuz I'm her I'm her conduit like what the tech world's actually like so honestly that's the 3:02:17 prim things like how am I going to explain this show Mom the money kind of something something like that as summarize uh Grace Alex thank you so 3:02:24 much for joining us today at AI Summit thank you thanks guys [Music] 3:03:17 returning to to the stage Nicholas Johnston good afternoon uh thanks for 3:03:24 sticking with us we are in the home stretch of our AI Summit today thanks for being here and special thanks again 3:03:31 to our partners who have made uh today possible especially meta they have an insulation in back where you can play 3:03:37 around with some of their very cool generative AI tools especially with imagery and before we go deeper on that 3:03:42 with somebody from meta I want to share a really quick uh video and I'll be right back so then the night a cute 3:03:48 flying dinosaur night so and he lives in a jungle with other animals I think they 3:03:55 should TR a big boat a big sailing sh ship but suddenly there's a storm oh no 3:04:02 they're in a storm who saves them a kind wizard with a funny hat say that to 3:04:08 Grandma yeah me SE I think she'll like her a lot 3:04:18 uh I'll be honest that stuff is wild uh so let's learn a little bit more and dig into that uh with metas VP of product 3:04:24 Connor Hayes Connor welcome to the axio stage how are you I'm good thank you um 3:04:32 so let's let's start a big picture here like what's the why like those like the process that I went through it met was 3:04:38 why to start putting these kinds of pools uh into products sure um I mean I think the thing I would start by talking 3:04:44 about on that topic is that AI at meta is not a new thing um we've had a 10year now investment in 3:04:51 a fundamental AI research lab called fair at the company um and actually a 3:04:56 lot of the stuff that you use every day in meta products like ranking videos or identifying content that should be taken 3:05:02 off of the platform is all powered by AI one of the big pushes that we've made in the last year though is to try to be a 3:05:08 lot more intentional about bringing uh AI powered products to the market like the ones that you just saw in this video 3:05:15 I think the why behind it is honestly that there's this big opportunity for us to take the things that people use meta 3:05:20 products for today and just make them a lot more useful and Powerful with this new technology so our whole strategy is 3:05:27 taking the things that you're coming to our products for applying this technology to it and then just making it more useful more powerful for you is it 3:05:34 almost like pulling AI more to the Forefront like AI sat behind so many things that we didn't really know about like I joke in The Newsroom like spell 3:05:40 check kind of an AI if you think about it but now it's pulling it Forward uh in 3:05:45 a way where I think users can see it I think so I think so and I think honestly um I you know I don't know how 3:05:52 much of a goal it is for people to to feel like the thing they're using as AI but instead to just feel like the thing 3:05:57 that they're using is really powerful and magical yeah um but you know this is 3:06:02 It's been a big Focus for us mostly because our products are so social people are coming together having conversations um the experience that you 3:06:09 just saw in this video is available in our messaging interfaces so things like that you can be send creating and 3:06:14 sharing images back and forth with folks in a group chat on WhatsApp or on messenger or something like that um and 3:06:19 we found that to be just a really great enhancement to the social utility that our products are let's dig it a little more on the user side of that like are 3:06:27 you picking up Trends on like who is using it and what they're using it for and how they're using it that's interesting yeah we are um so we we 3:06:34 expanded actually we launched um a new version of our our llama model in April called llama 3 which is also available 3:06:41 for for developers via open access all of these products are now powered by llama um when we made that change we 3:06:47 also expanded access to meta AI which is our assistant available across all of our apps and Rayband meta smart glasses 3:06:54 uh to 12 additional countries it's only available in English right now um but one of the the big learnings that we've 3:07:00 had since then is that the differences in how people are turning to meta aai and using it on a by country basis are 3:07:07 really interesting um so far you know for example we launched to a handful of countries in Africa where WhatsApp is 3:07:14 just a hugely powerful tool for communication those markets um and those 3:07:19 folks are turning to meta for more like research and information gathering and answering basic questions um whereas if 3:07:24 you look at you know how people are using it on Instagram Direct in the US it's much more creativity and entertainment and um creating images and 3:07:31 stickers with your friends so and it seems like we're littleit at the beginning of this now we finally put all these tools down to the market and we're 3:07:37 just beginning to learn like okay what are the use cases I think that idea about how like different even different regions are approaching it in a 3:07:42 different way um dig a little bit more into like what are the highlights challenges so far like we've been added 3:07:49 at the like as you said AI has been a thing for meta for 10 years but in this context It's relatively wish what what 3:07:56 are you learning the the Highlight for me and I've been with meta since since 2011 3:08:02 there's nothing like the feeling of taking a product that you really believe in and bringing it to more people in 3:08:08 more markets internationally um when we launched a lot of these products in September last year they were only 3:08:14 available in the US so that International expansion has been big highlight like I said it's English only 3:08:19 now but we're going to bring it to more languages and countries over the course of this year um just there are folks in 3:08:25 those in those markets who haven't used a tool like this before and seeing the uptake has been really uh encouraging um 3:08:31 I would say another one of the big highlights is how people are using the image generation tools that that you just showed off um when we dig into the 3:08:38 data to sort of understand these use cases that are really sticking that ends up being one of them um and a lot of it 3:08:44 is actually just for like pure entertainment um in that like like sort of chat or group chat use case invoking 3:08:50 meta AI to create images you know with your friends and sort of riffing on that got like people figuring out how to 3:08:55 begin incorporating their life it's start it's you start started with like as almost as a play thing and then you get as you figure it out maybe it'll 3:09:01 take reach that that next step you think yeah actually I mean it's a a great lead into one of the things that we're also 3:09:07 seeing is you will be introduced to this this technology through something that's like fun and social like that but then 3:09:14 also learn that it can be used for you know learning or getting things done so if You observe your friend creating an 3:09:20 image in a group chat maybe the next time you might ask for a recipe or recommendations on where to go for a 3:09:25 night out in the same context is there a flip side of that are there challenges that have risen up things that have seen that have made it a little more 3:09:30 difficult than you would have thought definitely a lot of challenges um you know AI does not come without its 3:09:35 challenges I think probably the the the biggest industrywide challenge that we're also facing is something that you 3:09:42 would call like alignment or controlability of the model so um you know in a product like meta AI we have a 3:09:48 set of content standards that we of topics that we want it to engage on and not engage on or you know sort of the 3:09:55 voice and tone of the assistant that you want to maintain consistently um and just dealing with keeping the model 3:10:00 aligned to the sort of rules and tone that we've given it has been a big challenge um probably another one I 3:10:07 would say um has just been as we've as we've gone to these additional countries 3:10:14 with more features learning what's working and not working in real time and being able to react to that so a big uh 3:10:20 launch in this in this April set of launches that we did that many people here have probably experienced is we put meta aai into the search bar across our 3:10:27 apps on Facebook Instagram messenger WhatsApp now when you're searching you can also search via meta aai um those 3:10:33 are really highly trafficked Sur surfaces across those apps that's like hundreds of millions of people a day so 3:10:39 anytime you make a change to a thing that hundreds of millions of people are used to using a certain way you're going 3:10:44 to run into some challenges feedback is going to come fast it has come very fast uh and we're working through fixing fixing a lot of it yeah uh so what's 3:10:51 next what's coming down the pike what are you excited about what's in the secret lab that you want to tell us about the first time ever the there's 3:10:57 cauldrons of bubbling uh uh the I think the biggest thing is this International 3:11:02 expansion bit that I pushed on before um we're working really hard to make the the product available in more countries 3:11:08 and languages honestly the the biggest challenge there is the alignment bit that I mentioned before because the 3:11:14 things that you have to do to make a model align to your stand standards in English in the US are actually really 3:11:19 different than if you go to Tagalog in the Philippines for example there are cultural norms and differences um 3:11:25 there's you know localized issues that you want to make sure that you're sensitive to so that's something that we're taking our time with um but but 3:11:32 we're really excited about like the next big thing is like figuring out how to make this Global truly Global exactly and then you know we we launched meta AI 3:11:38 as well as mult multimodal understanding on the Rayband metas I have them on now you're wearing I could ask them what I'm 3:11:44 looking at it I don't know if that it would know you because we can't identify individual people um but so that has 3:11:49 been really popular but again getting that meta AI integration on the rayb bands available in more countries is a big Focus for us as well uh all right so 3:11:56 I could ask you everyone knows like to end on one fun thing we could end on one fun AI thing but I know you're a Celtic Super Fan and the NBA Playoffs true 3:12:03 final start tomorrow maybe in a room of Nicks fans so I don't know Well w wamp 3:12:09 uh give us give us your I'm going to get run out of town here uh give us your prediction Celtics in how many games oh 3:12:14 my God this is hope will this played back when we l on the there's a lot at stake here for the Boston Celtics I just 3:12:21 want know tell a Nick's CR how quickly the I'm going to go on the record Celtics and five uh it's going to it's 3:12:27 going to be a Calk Fair all right you didn't lose the crowd completely Conor thanks so much for being here huge thanks to meta uh for making today 3:12:33 possible and IA is coming back for a big finish awesome thanks everybody thank you than you 3:12:39 [Music] [Applause] 3:12:47 give it up again for Ena [Music] 3:12:53 freed so our next guest was listed on times top 100 for most influential 3:12:59 people to know when it comes to AI you may know her as a very vocal advocate for privacy in an era where the 3:13:06 technology seems to have made privacy seem like an Antiquated notion I'm really glad we're ending today although 3:13:14 we have one demo after that but this conversation is so important I'm fascinated by the technology I think 3:13:21 there are a million interesting uses good uses of the technology but it's 3:13:27 really important that we get this right and I'm really hope that we have a really thoughtful conversation um that 3:13:34 leaves you all with a lot to think about so please help me welcome the president of the Signal Foundation, Meredith Whitaker 3:13:42 [Music] thanks Mida um I mostly want to talk about AI but I'm actually going to 3:13:49 start with something that's slightly orthogonal um which is um signal has 3:13:55 been the subject of a pretty intense campaign for those who don't know signal you should how many people use signal 3:14:01 how many people use signal okay well by the end of this I think we'll double that um it's it's an encrypted 3:14:08 confidential messaging system that's actually encrypted um and you've been 3:14:13 subject to a pretty intense campaign from telegram one of the competitors but also a lot of people who stand to 3:14:20 benefit from weaker security why should the public care well I think everyone 3:14:27 cares about their privacy if you close your eyes and imagine every email you've ever sent dumped in a database 3:14:33 searchable by everyone you know you feel that right we all care we 3:14:40 all have stakes in this and when we look at this from a geopolitical perspective 3:14:45 we also recognize as scholar ERS of history or even people who are observing the Dynamics in our politically volatile 3:14:51 time now that those in power often constitute their power via information 3:14:57 asymmetry by knowing more about their adversaries or their subjects and using 3:15:03 that to their advantage to oppress to undermine to have an advantage at the negotiating table I think anyone who 3:15:10 spent some time in the Snowden documents just knows exactly how powerful these forms of mass surveillance can be when 3:15:17 in the hands of an interested government so I don't think I have to belabor the 3:15:22 point that privacy is important and that privacy in the context of a tech 3:15:29 ecosystem that currently constitutes itself via Mass surveillance via collecting huge amounts of data using it 3:15:36 to sell ads or train AI models perhaps um provides a really significant attack 3:15:44 surface when we're talking about that combined with some of the will to power ... some of the will to surveil and the potential to weaponize that type of centralized surveillance data so are you 3:15:56 guys with me and the big why it matters is you know we used to have a fair amount of our privacy came through obscurity it just before technology you just couldn't listen to everyone's 3:16:07 conversation that's not protecting us anymore literally for hundreds of thousands of years most of human 3:16:13 development most things were by default private right you talk to someone maybe 3:16:18 someone was eavesdropping but they were someone you knew they were in your little town your Hamlet your fight them 3:16:24 whatever and that wasn't logged in a database and then sort of read with some OCR technology immediately searchable by 3:16:32 a centralized Corporation in the jurisdiction of the US which is right now in what my favorite euphemism right 3:16:38 now politically volatile times right it's a very very different Paradigm and I think over the last 20 years what 3:16:44 we've seen is an industry grow up around claims of human benefit around claims of 3:16:50 scientific progress that we're just now recognizing is built on some pretty 3:16:56 toxic foundations on the centralization of surveillance on the centralization of technological power 3:17:04 and most of the companies that have that kind of Monopoly status that have sort of built themselves as you know the the 3:17:12 surveillance Giants the big tech companies whatever you want to call them are jurisdiction in the us so I want to 3:17:18 bring it now into the generative AI moment we're in again I feel like today is offering a reality check of where we 3:17:25 are um one of the things that is undoubtedly true about this moment is most of the technology is being provided by a few companies you've talked a lot about why that probably isn't a good thing where are we in this moment in terms of the concentration of power within generative AI yeah I I I love this question and I can't answer it without reference to history because I think there's a really important question that underlies this that we don't ask often enough and oftentimes maybe that's because we're embarrassed we're not sure like does everyone else know this I don't know this but that question is what is AI right AI in my view having you know gone through the archive spent you know over a decade studying this is more or less a marketing term it was invented in 1956 by the cognitive and computer computer scientist John McCarthy who's now named a father of AI and in subsequent interviews he's super clear about why he coined the term there were two primary reasons the first he wanted to exclude an academic rival a man named Norbert weiner who had invented the term cybernetics and at that time the field was constituted under cybernetics he didn't want to invite this guy he wanted to be the father of his own thing not a disciple to Norbert pretty common academic will the second reason which is even more common in Academia is he wanted grant money because at that time we're just following World War II this is the Cold War era where dreams of omniscience in the face of nuclear Russia were motivating a huge amount of funding into computational Technologies to try to have that cold war Advantage so he was looking for DARPA funding and he coined the term artificial intelligence and it kind of stuck but over the subsequent 70 plus years that term has been applied to many many different technologies that really don't look anything like each other so I think we can use that to understand it's very flexible it's kind of un you know it's a it's a floating signifier to be a bit theoretical about it and then the question becomes why are we obsessed 3:19:31 with it right now why did it emerge in the last 10 years as the big new thing why did it have its Resurgence and to 3:19:39 understand that we need to recognize at the base of this technology is this Monopoly surveillance business model are 3:19:46 the companies that through the 2000s established themselves the Googles the metas the microsofts as platform 3:19:53 monopolies as Cloud monopolies were able to collect a huge amount of data and 3:19:59 build up huge powerful infrastructure to feed these businesses so you most of 3:20:04 this was advertising kind of you know calibrating ads and engagement you know 3:20:10 there's also the cloud business model but effectively the AI we're talking about now is a derivative of that 3:20:16 business model it takes huge amounts of data collected by these companies and we know like data is scarce everyone's sort 3:20:21 of you know trading you know trying to broker deals with different news organizations or Reddit get as much data 3:20:27 as you can to trade your AI and then it was the companies that had the data and had the powerful computational 3:20:33 infrastructure to actually run this training and access market so you know none of that means AI isn't useful but 3:20:41 it does mean it is very contextually it it is in the context of this surveillance business model and it 3:20:48 is favoring the mon the the big Tech monopolies that establish themselves so 3:20:54 that they do have that infrastructure they do have that data it's not so much 3:20:59 a product of scientific progress it is a product of a recognition you could do new things with old algorithms when you 3:21:06 had compute and when you had data that again very few actors have and how are those companies doing for example I was 3:21:13 in Seattle a couple weeks ago Microsoft showed off this new tech technology for the PC called recall where it's going to 3:21:19 let you recall anything you've ever seen because it's taking constant screenshots 3:21:25 and saving that is that a good thing for society oh God no 3:21:30 um I play softball in case you don't know yeah is that um do we love this no 3:21:36 um I mean I think we should be horrified right at least those of us who have a you know a bit of depth of understanding 3:21:42 of what this Tech actually does and the track record of these companies and the political environment in which they're 3:21:47 operating in right you know I I do get this question as the president of signal someone very devoted to privacy why do 3:21:54 you spend so much time talking about AI why do you continue engaging in academic 3:21:59 scholarship around Ai and recall I think is a very good example that the vend 3:22:04 diagram between privacy and AI is actually a circle because what recall is 3:22:10 in my view is a desperate attempt to find a market for Capital intensive 3:22:15 technology that need a return on investment so Microsoft is shoving this you know what 3:22:22 is effectively spyware into their new um and into their new devices into their 3:22:28 new version of Windows and already the security reviews are horrifying recall 3:22:34 as it is constituted now is on by default so you have this spyware taking snapshots of everything you do then 3:22:40 subjecting them to ocular you know basically AI that scans those and transcribes those into a data base that 3:22:47 anyone with access to your computer whether they get that remotely or with physical access to your device can then 3:22:53 read that is an incredibly dangerous Honeypot for hackers hostile nation states anyone who gets access to your 3:22:59 machine and that will be deployed again opt out which means it's working on 3:23:04 everyone's device you know without most people knowing because we know very very few people change the default settings 3:23:11 so this is an incredible I would say you know the disrespect inherent in making this move 3:23:17 the actual abdication of responsibility for a company that is providing core infrastructure for billions of people 3:23:25 around the world to subject them to this kind of danger and for me the offense is really this is you know people run 3:23:32 signal on their desktop people for whom privacy and security is a life or death 3:23:38 issue and do you think some of this comes from the fact that there are a lot of basically most of Silicon Valley is 3:23:45 run by mostly white mostly men who maybe don't understand how the information might be 3:23:53 used against them their own data I mean yes I think people who haven't faced 3:23:59 many risks or have a you know blood level understanding of marginalization and what it means to be subject to you 3:24:06 know those with power over you make really bad decisions about this fairly frequently but I also think there is a 3:24:12 sort of group think around AI where people aren't pausing to actually differentiate between where it are large 3:24:21 models that recognize patterns in data useful because there are places right this is not useless technology and where 3:24:28 do we need to leave it alone and leave money on the table and that question is not being answered and particularly in 3:24:33 the context you know it costs hundreds of millions of dollars billions of dollars to train these models so the capital you have to throw at training 3:24:41 and then calibration and then inference as well is huge and so there is deep pressure from companies that are 3:24:47 basically promising God and delivering email prompts to make some return on investment on this technology and I 3:24:54 think there's you know this is a sign for me of desperation it's a sign of Microsoft just you know shooting 3:24:59 themselves in the knee and again they are snapshotting signal messages on device that is an incredibly profound 3:25:07 violation of trust and you know you've just um you know I I think we really 3:25:12 need to recognize just how serious you know kind of hijacking people's 3:25:19 trust and their you know digital environments um you know how serious 3:25:25 that hijacking of trust really is and and think you know kind of zoom out and think what other what other powers have 3:25:32 we seeded to these companies that we really need to take back and you know regain the ability to deliberate and 3:25:38 direct more democratically given that that is where the power is given that that's where the 3:25:44 technology lies is in these four companies given that they're investing billions is there a different way you 3:25:50 think this story can play out and how would that happen well lovely that you 3:25:55 asked because signal is actually a nonprofit um and we are a nonprofit you 3:26:01 know not because this is a nice little Charity but because we looked at the cold hard business model of tech and 3:26:08 realized that if we were a four profit it is very likely that we would be pushed to erode our privacy guarantees 3:26:15 in an industry where collecting selling making use of personal data is the primary economic 3:26:22 driver so I think there are certainly other models of tech there are certainly other models of organizing our economy 3:26:29 of organizing our social relations all of those you know kind of fit together but I think we need to take very 3:26:35 seriously the fact that you know we're in a politically volatile time these companies you know serve 3:26:43 whatever government they're jurisdiction in they have have data and the power to 3:26:49 control the infrastructure across our lives and institutions that has never been seen in human history you know this 3:26:55 would the the stacei would cry with you know absolute Glee if they recognize 3:27:01 just how much access they could be given through these Technologies how much insight into people's individual lives 3:27:07 how much power to infer highly personal things about people and communities they 3:27:13 would have if they deployed these Technologies and right now those in the hands of for-profit companies that will act like for-profit companies putting 3:27:20 revenue and growth above other things because that's how they're constituted and that is not a healthy way to make 3:27:26 such significant decisions about you know not only us but you know Global 3:27:32 institutions governments and societies but I did hear in what you say also a cautionary tail to those who look too 3:27:37 far to governments to regulate this that their interests Also may not be aligned How concerned are you that in in our 3:27:45 efforts to try and counterbalance the giant tech companies we hand over too much power for example I think you've 3:27:50 spoken about the dangers of a Tik Tock ban in how for example if we have another Trump Administration how that 3:27:57 might be weaponized yeah I mean I think we need to be very careful about assuming you 3:28:03 know a kind of almost like colloquial Superman story that we're just creating tools that the good guys will use on the 3:28:10 bad guys right and my my partic you know the Tik Tok band was clearly kind of politically weaponized and my my 3:28:17 argument there was that you know giving the executive branch in the context of a 3:28:22 you know authoritarian looking regime the power to declare based on sort of you know fairly flimsy evidence at least 3:28:28 what was presented publicly that you know some entity is I think it was foreign foreign adversary controlled was 3:28:34 the term is an incredibly flexible tool and you know post 911 we saw similar 3:28:42 Powers misused over and over again so you know I think we I think we need to recognize that you know the the 3:28:51 significant political will that would accompany kind of getting access to surveillance data 3:28:57 getting access to the information platforms that are currently controlled by five companies four of whom are 3:29:03 jurisdiction in the US so you know Tik Tok YouTube X Etc um and that that is 3:29:08 you know traditionally governments have sought exactly that kind of power through you know control over news media 3:29:14 control over more analog forms and and right now it is centralized and you know 3:29:19 much easier to influence and access than those previous iterations and spans 3:29:26 globally so you have you know most of the world Reliance on five companies four of whom are jurisdiction in the US 3:29:33 to you know to understand our shared reality to access our shared media ecosystem well unfortunately on that 3:29:40 light note we are going to have to leave it there um but I encourage you to keep 3:29:46 following uh Meredith's work um she's been working in this space since before most of us were using the term AI she 3:29:52 led the Google walk out uh to pay attention to project Maven which um if 3:29:58 you've been following the news Ai and warfare is a whole another subject we didn't get to get into today but please 3:30:03 keep following Meredith's work thank you so much thank you Ena thank you all so I will be back in just one second 3:30:10 we're going to set up for our very last demo and then I'll be right back 3:30:16 [Music] 3:30:40 [Music] oh o 3:30:48 [Music] 3:31:22 [Music] 3:31:36 welcoming back Ena freed all right thanks for bearing with me uh before we get to our last demo I 3:31:43 wanted to thank um a few people that made today happen first of most all of you for being here all the folks on the 3:31:49 live stream for watching thank you um thank you to the Altman building team really quickly because we did get this 3:31:56 question no relation to Sam it's actually the now defunct be Altman department store and its only Tech tie 3:32:03 is that in the uh marvelous Miss masel show she takes a job at what was then be 3:32:10 Alman so in case you were wondering I looked it up for you Wikipedia thank you 3:32:15 I want to thank our sponsors IBM and meta for their support we couldn't do this without them our catering team our 3:32:22 production vendor Pro Show and of course our amazing axios events team thank you 3:32:27 don't forget to stick around after this we're going to have some bites beverages I'll be down there it'll be great look 3:32:33 forward to catching up hearing what you thought and if you aren't subscribed to our newsletter the debates the 3:32:38 discussions that you've heard we really try and have that every day I think it's a real luxury that I get to write about 3:32:45 AI every day and pick one topic and say here's something we should be thinking about if there's stuff you think we 3:32:51 should be talking about that we're not drop me a note I'm just en aos.com if 3:32:56 you don't get the newsletter you just go to AI plus. axios.com we'll get you hooked up um and now for our closing 3:33:04 segment I want to introduce our next guest who heads up the company gecko robotics um their robots are not the 3:33:10 sort of humanoid type robots they're not the Boston Dynamics dogs that you usually think of they are used to gather 3:33:17 data in places that are hard to reach for humans robot humans dogs robot dogs 3:33:23 um think like a loose wheel on a train or um creating digital twins to gather 3:33:29 data to fix damage in real time here to explain where AI fits into this is Jake 3:33:34 luaran the CEO and co-founder of gecko robotics Jake 3:33:40 [Applause] [Music] 3:33:45 so yeah so we're not talking when when people think robots these days for better or worse you know there's a lot 3:33:50 of humanoid robots out there um but it seems to me that's a bit of a trap in the sense of the whole point of building 3:33:57 a robot is you can build it to the task you want to do that's kind of what you've done we're going to look at a 3:34:03 couple photos and you're going to tell us what that picture is and what's wrong with it so if we can uh cue these two 3:34:11 photos all right so here we have a bus crash yeah so the headquarters of P of geek 3:34:18 robotics is in Pittsburgh so this is actually the front Hollow bridge that collapsed before uh President Biden uh 3:34:23 showed up that day um so a big topic was infrastructure and you're right yeah our 3:34:29 our robots don't dance to cardi b or if they do it's not really that interesting um so like the uh the the this like idea 3:34:36 of Robotics that can do really cool things um I that's what got and got me 3:34:42 fascinated with the robotic sector but then I started looking deeper trying to understand what are the unique 3:34:48 things that robots can do to actually impact real world problems something like this for example is a real world 3:34:55 problem um but it doesn't necessarily have like the uh draw to um The Tech 3:35:02 Community or other roboticist or um folks in software or AI um it's hard to 3:35:08 solve a problem when you don't like hear a lot about it or aren't affected by it 3:35:13 but these kinds of moments are happening at a much faster and more increased rate 3:35:19 critical infrastructure like this bridge they're failing and we have a big 3:35:25 problem a lot of these Bridges things like that really old um we know they're past their design life but we don't know 3:35:32 which ones are at risk and that's what you helped show if we can cue the next photo um part of the challenge is to 3:35:39 inspect Bridges pipes you know Wells tunnels it typically requires this which 3:35:45 is time consuming expensive dangerous what can we do with robotics that we 3:35:51 can't do by sending a human up there great question so I call this Joe he's 3:35:57 on a rope and he's probably suffering from a pretty intense wedgie because he's been up there for a while um this 3:36:04 was uh this was the image that captured my attention about 10 or 11 years ago 3:36:10 started the company at of college uh college dorm I got a chance as an intellectual engineer to go to a power plant nearby where went to college in 3:36:16 Western Pennsylvania and um I got to see that a power plant was shut down and failed 3:36:22 because of a pressure vessel explosion so you know blackouts and brown outs have been occurring uh at a 2X rate for 3:36:29 the past 5 years um why it's happening is that we increase the demand on Old infrastructure that was built in the 3:36:34 Cold War and it's beginning to fail especially when loads increase and the best way to stop it from happening is 3:36:40 sending folks like this um up with um tools that are really similar to Medical 3:36:45 tools Ultrasonics is what this individual is using to test the structural Integrity of a really important piece of infrastructure and um 3:36:53 this person at the power plant I went to um had died the year before doing this job it's very dangerous and 3:36:59 unfortunately you're just not really able to capture a lot of information and data about the physical world and this is a problem it's really great to solve 3:37:06 um and use technology and artificial intelligence to do incredible things um but if you don't actually have the 3:37:13 underlying data sets and the first order data sets in particular then you have a hard time solving problems like how do 3:37:18 you like real problems things like how do you ensure that the infrastructure that we all depend on every day is going 3:37:24 to be there for us for the next 10 20 30 years especially as you begin to transition talk about transitioning to 3:37:30 uh renewable energy sources it's really important to make sure that you're reliable and so what the robots can do 3:37:35 is they can collect more data more often and help you not just see visible damage 3:37:42 but kind of get a sense for when things are going to go wrong so you know there's limited budgets you know if we 3:37:48 could build every new bridge tomorrow we still couldn't afford it we still wouldn't have enough people to build it 3:37:53 we have to prioritize and basically your robots help us do that right I'm not sure do we have a video I can't remember 3:37:59 if we have a video we do we do um and I think it's important some numbers for you um you know we uh um corrosion costs 3:38:07 um us about like 3.6% of global GDP every year um we're at like a D um um 3:38:14 grade as relates to our infrastructure um it's about a $4.5 trillion dollar problem to get us like to a b um and and 3:38:21 I think like these are all like really scary numbers and there's like 177,000 bridges in New York and only six are not 3:38:28 in need of immediate repair this is like a very doomsday scenario but what I'm trying to communicate is that you need 3:38:33 information and data to begin to solve one the prioritization of those problems but two be to understand how do you um 3:38:40 not just maintain but then build smarter pieces of infrastructure into the future so in College I built a wall climbing 3:38:46 robot to go out and gather information and data to then predict when things were going to fail this power where are 3:38:51 your robots in use today and how many are out there so after 10 years of gathering over five a data on over 3:38:58 500,000 of the world's most critical assets um uh we've been able to not just build robotics like this um and leverage 3:39:07 um drones and walking dog robots and and fixed sensors we've been able to collate all of that into a centralized software 3:39:13 called canver that manages um and is a source of Truth um for the biggest 3:39:18 energy companies whether it's oil and gas or power um the Air Force and the Navy both um domestic and Allied um the 3:39:28 uh the manufacturing sector the public infrastructure sector on dams and bridges um you know the the it's a far 3:39:34 we're far we we have penetrated into a bunch of different sectors that of suffer from similar problems in the 3:39:40 physical domain um and that aggregation of all this information is actually allowed for us to make some pretty 3:39:47 interesting conclusions as relates to what will fail and when but also how do you ensure that you can get more out of 3:39:52 the infrastructure um to solve for the critical problems for a customer like how do you increase production of energy 3:39:58 while also decreasing emissions levels um that's a literal problem that we're solving for some of the biggest 3:40:04 companies in the world and if we have that video Let's play it and then rather than just talking about it I think it'd 3:40:10 be fun to see the robot in action don't you guys all right so if we have the video Let's play it if not it sounds 3:40:16 like we don't no problem you don't need a video we've got this live and I'm going to control it with an Xbox 3:40:22 controller is that right yeah yeah let's not over complicate things all right now I'm terrible at video games but one of 3:40:28 the things that's nice about this is this isn't just like it's not a humanoid 3:40:34 robot this isn't the world's fastest robot this isn't going to do a race but it is pretty cool the way it can move 3:40:41 yeah it is it's going to connect to Bluetooth right now so right now it's searching for the um it's searching for the um the pad 3:40:48 right now so while that's happening what I'll describe for you and actually um let me hand it let me hand it over to to 3:40:54 will to connect it so what what we're going to do is you'll see it kind of looks weird so we'll I'll desra this is well everyone say hi to well hey Will 3:41:01 good job will so this is the robot um it looks so just like a CAT scan on a bot 3:41:06 and when you're doing a when you're going in for a Health Scan you have a suite of Ultrasonics up here that are basically doing a sonogram of the 3:41:13 physical structure and while it's doing that it's using cameras it's using Lars it has IMU on board and we're we're not 3:41:20 just Gathering the information the data see lights um about the structure we're actually mapping out and creating a 3:41:25 digital twin because as you know you want to be able to create a historical record um that can extend the useful 3:41:31 life of the critical infrastructure that can help it perform better and if you have if you can map on a digital twin um 3:41:38 and the Health Data then you can begin to pull in more sources of information and so this is what was really interesting with the optimization 3:41:44 problems we're begin to solve so um so we're Gathering a couple different kinds of information from the ground level and 3:41:51 um and then we Port that into the centralized software which is Canever so we'll get that moving for in a in a 3:41:56 quick second here and if it doesn't happen I mean the important Point here is we don't have to run these robots 3:42:02 they are but what it's doing is it's giving us more confidence that that old bridge that we're going over won't uh 3:42:09 won't collapse on us well exct I'll have to save for next time playing with the controller but thank you so much Jak H 3:42:15 around I got a couple uh last things uh feel free to have a seat um of so this been fascinating thank you again 3:42:21 everyone for coming out today we're going to round out the program with a few fun things that you're going to want 3:42:26 to stay for on the screen we should have the winners of the photo contest and 3:42:31 then I have one more surprise can we cue that 3:42:39 slide all right there we go so um we've got some dancing zebras we've got a 3:42:45 biome and a rocket dog so um if any of those are your Creations uh if you come 3:42:52 to the uh check-in desk um we will hook you up with your prizes and then I have 3:42:58 one more thing just need those 3:43:06 down oh was waiting for Ena to leave there's our robot so it's doing again I'm not 3:43:13 good at video games Jake Jake's doing it better than I could um so before she 3:43:20 left and you're going to want to put your phones down for this part uh before she left uh Maria was nice enough to 3:43:26 sign some tennis balls for us and I'm going to just throw these into the crowd 3:43:32 so um I hope axio has some very good insurance if you want a tennis 3:43:39 ball I have a pretty good arm I pitch softball overand 3:43:53 tennis is not my sport there we go 3:44:09 oh it's a trick [Applause] 3:44:16 yeah I see all the hands what I'm not having any success is getting these balls 3:44:22 out all right there we 3:44:31 go nice throw all right who thinks I can get it all the way to the back without 3:44:37 hitting the [Applause] 3:44:43 lights with that note let's all have a drink cocktails mocktails and some bites 3:44:50 [Applause] [Music] 3:45:36 [Music] 3:45:55 [Music] 3:46:01 [Music] 3:46:20 [Music] 3:46:42 [Music] 3:47:01 [Music] 3:47:07 [Music] 3:47:13 [Music]

SITE COUNT