1 00:00:00,240 --> 00:00:05,028 [MUSIC] 2 00:00:05,028 --> 00:00:10,461 Sustainability unwrapped a conversational podcast about responsibility, 3 00:00:10,461 --> 00:00:15,397 ethics, inequalities, climate change, and other challenges of our 4 00:00:15,397 --> 00:00:19,920 times where science needs practice to think about the world, and 5 00:00:19,920 --> 00:00:24,307 how to make our society more sustainable one podcast at a time. 6 00:00:24,307 --> 00:00:24,922 [MUSIC] 7 00:00:24,922 --> 00:00:29,553 >> Hello everyone, and welcome to a new episode of the sustainability 8 00:00:29,553 --> 00:00:32,975 unwrapped podcast, my name is Robert Chiquita. 9 00:00:32,975 --> 00:00:38,091 And I am an assistant professor at the Hong Kong School of Economics and 10 00:00:38,091 --> 00:00:41,819 the Department of Marketing, and I do research and 11 00:00:41,819 --> 00:00:46,340 I teach on all sorts of topics related to digital marketing. 12 00:00:46,340 --> 00:00:52,300 And today I'm happy to be joined by two wonderful guests to talk about 13 00:00:52,300 --> 00:00:59,520 subjects on artificial intelligence AI from a couple of different perspectives. 14 00:00:59,520 --> 00:01:05,361 So it's my great pleasure to introduce a council Malik and Mao Rego and 15 00:01:05,361 --> 00:01:10,422 apologize in advance if I completely butchered your names. 16 00:01:10,422 --> 00:01:15,011 But I would like to give you the floor too briefly introduce yourselves. 17 00:01:15,011 --> 00:01:20,028 Tell us a bit about your your background and how you ended 18 00:01:20,028 --> 00:01:25,919 up being interested in topics related to AI or or working with AI. 19 00:01:25,919 --> 00:01:30,240 So a contra I'll give you the first word. 20 00:01:30,240 --> 00:01:33,092 >> No you're back on track with the name. 21 00:01:33,092 --> 00:01:34,820 So you're all good for today. 22 00:01:34,820 --> 00:01:36,632 We've trained this well we'll be fine. 23 00:01:36,632 --> 00:01:41,022 But yeah my name is Khan to I am a data scientist in Melbourne Australia, 24 00:01:41,022 --> 00:01:44,679 living the land of the free at the moment where it's no COVID 25 00:01:44,679 --> 00:01:49,740 we'll just I'm not going to say that again because I feel like I'll jinx it. 26 00:01:49,740 --> 00:01:54,388 But yeah I worked as a data scientist and here I'm also a Microsoft AI MVP. 27 00:01:54,388 --> 00:01:58,848 And kind of how I got into the whole world was I did a degree in math's and 28 00:01:58,848 --> 00:02:01,270 statistics and actuarial science and 29 00:02:01,270 --> 00:02:05,729 kind of what I came out of that knowing is I don't really like finance or 30 00:02:05,729 --> 00:02:09,888 statistics that much in like very technical statistic levels and 31 00:02:09,888 --> 00:02:14,840 didn't really enjoy the actuarial science too much in terms of doing a job. 32 00:02:14,840 --> 00:02:17,741 But I did like the data stuff we did I did like the computer stuff we did. 33 00:02:17,741 --> 00:02:23,821 So I ended up coming into a grant program doing data science as consulting and 34 00:02:23,821 --> 00:02:27,245 it's just spiral since then and here I am. 35 00:02:27,245 --> 00:02:28,360 >> All right thank you. 36 00:02:28,360 --> 00:02:30,847 And then we also have morrow. 37 00:02:30,847 --> 00:02:31,695 Hi Morrow. 38 00:02:31,695 --> 00:02:32,562 >> Hi. 39 00:02:32,562 --> 00:02:36,026 Yeah so how I what I do and how I cannot here so 40 00:02:36,026 --> 00:02:39,981 I'm a designer at google research or google AI. 41 00:02:39,981 --> 00:02:45,691 And I ended up there because I got a job there. 42 00:02:45,691 --> 00:02:47,998 I think that's the first element like I did. 43 00:02:47,998 --> 00:02:51,597 It was not a choice that I made like consciously saying 44 00:02:51,597 --> 00:02:56,112 I want to work more with AI machine learning was just a fact of like going 45 00:02:56,112 --> 00:03:00,710 through different projects and happening to ended up gathering enough 46 00:03:00,710 --> 00:03:06,340 experience to ended up in a position inside of the research unit at google. 47 00:03:06,340 --> 00:03:08,840 Yeah. 48 00:03:08,840 --> 00:03:13,240 >> All right so as you might see that if you had a look at the title of 49 00:03:13,240 --> 00:03:18,079 today's episode is kind of AI from different perspectives right? 50 00:03:18,079 --> 00:03:22,001 So from data science to design and many places in between. 51 00:03:22,001 --> 00:03:26,482 So you can kind of understand now once you once who our guest stars, 52 00:03:26,482 --> 00:03:31,127 you can understand what the idea behind today's conversation is but 53 00:03:31,127 --> 00:03:33,305 let's maybe just dive into it. 54 00:03:33,305 --> 00:03:37,457 So I had I had a talk with my dad this this weekend in which I was 55 00:03:37,457 --> 00:03:42,041 complaining about my work clothes that seems to be never ending. 56 00:03:42,041 --> 00:03:47,407 And he said something funny which was can't you just program some AI algorithm 57 00:03:47,407 --> 00:03:52,379 to do your work for you or some of your work for you so you get some time off? 58 00:03:52,379 --> 00:03:56,049 And I got completely stumped because I realized okay so 59 00:03:56,049 --> 00:03:58,771 that's what my dad thinks thinks AI is. 60 00:03:58,771 --> 00:04:01,616 And then we ended up having quite a long conversation in which I was trying 61 00:04:01,616 --> 00:04:02,140 to explain. 62 00:04:02,140 --> 00:04:04,703 Well dad actually that's not what AI is. 63 00:04:04,703 --> 00:04:09,416 So I would like to to kind of open the discussion by asking you 1st and 64 00:04:09,416 --> 00:04:11,411 1st foremost, what is AI? 65 00:04:11,411 --> 00:04:16,258 In a layman explanation or how would you explain to my dad what AI is. 66 00:04:16,258 --> 00:04:20,473 And do you have any examples or any ideas of what is good Ai or 67 00:04:20,473 --> 00:04:22,288 what can AI actually do? 68 00:04:22,288 --> 00:04:25,456 >> I think I was trying to I asked about this and 69 00:04:25,456 --> 00:04:30,296 was like how do I even quantify this because AI has just become this 70 00:04:30,296 --> 00:04:35,136 like buzz word that people just love using it like AI the end all and 71 00:04:35,136 --> 00:04:40,864 be all that will solve all of our problems kind of like well do your job for you. 72 00:04:40,864 --> 00:04:42,828 And I think I remember trying to explain to my sister, 73 00:04:42,828 --> 00:04:44,151 it's like the same thing, right? 74 00:04:44,151 --> 00:04:47,021 You try and explain it to your family who don't know what the hell you're doing 75 00:04:47,021 --> 00:04:47,641 any single day. 76 00:04:47,641 --> 00:04:52,190 But they kind of like you do something, I think you have a job to do something. 77 00:04:52,190 --> 00:04:55,317 But I think the easiest way I think of ai anyway and I might be wrong. 78 00:04:55,317 --> 00:04:59,427 I mean that's just the way I can see it is just the AI is the extra 79 00:04:59,427 --> 00:05:01,577 brain power that I don't have. 80 00:05:01,577 --> 00:05:05,340 And I can't leverage like it's the extra stuff I can't get from coffee. 81 00:05:05,340 --> 00:05:09,591 That's what AI is to me it will do the exact same thing that I can sit down and 82 00:05:09,591 --> 00:05:14,530 do if it's something like computer image, computer vision where it looks at an image 83 00:05:14,530 --> 00:05:19,052 and it's like, okay, this is a dog, I can do that AI just does that at a scale of 84 00:05:19,052 --> 00:05:22,456 about 40,000 images in two seconds kind of a thing. 85 00:05:22,456 --> 00:05:25,131 It's just that extra brainpower from my opinion. 86 00:05:25,131 --> 00:05:29,541 Anyway, I've seen around and what the use cases I've seen 87 00:05:29,541 --> 00:05:32,328 >> All right Morrow what is AI to you 88 00:05:32,328 --> 00:05:32,960 then? 89 00:05:32,960 --> 00:05:40,100 >> I like what your father said because it goes really in the direction of how I 90 00:05:40,100 --> 00:05:45,742 understand the whole thing is actually I would say AI is for 91 00:05:45,742 --> 00:05:51,961 cognitive work stands for cognitive work as machine to stand for 92 00:05:51,961 --> 00:05:56,130 for labor for for arts and crafts, right? 93 00:05:56,130 --> 00:05:58,921 So if I want to make tables right? 94 00:05:58,921 --> 00:06:03,897 And I can I can really sculpture table out of wood with my hands I 95 00:06:03,897 --> 00:06:05,919 can produce that right? 96 00:06:05,919 --> 00:06:09,381 And I'm gonna take I don't know how much time like three weeks to produce 97 00:06:09,381 --> 00:06:09,961 one table. 98 00:06:09,961 --> 00:06:11,516 I'm gonna cost a lot of material. 99 00:06:11,516 --> 00:06:13,740 I'm gonna be heavily expensive. 100 00:06:13,740 --> 00:06:17,564 But I can also produce that in a factory, right? 101 00:06:17,564 --> 00:06:23,111 And that's what AI kind of stands for right now in terms of business and 102 00:06:23,111 --> 00:06:29,130 and industry pretty much turning cognitive work work that like identifying 103 00:06:29,130 --> 00:06:36,440 pictures for example or solving solving problems bigger scale and automated. 104 00:06:36,440 --> 00:06:39,352 So the same relationship that you have 105 00:06:39,352 --> 00:06:43,361 would have with industrial revolutions are today. 106 00:06:43,361 --> 00:06:47,291 >> And I think probably there there are quite a lot out there who would who would 107 00:06:47,291 --> 00:06:48,801 like to use this term of now. 108 00:06:48,801 --> 00:06:53,207 We are living in this data revolution and the kind of algorithm revolution and 109 00:06:53,207 --> 00:06:57,696 we move from from industrial revolution to information revolution and so on. 110 00:06:57,696 --> 00:07:00,741 So again quite a lot of buzz words thrown around. 111 00:07:00,741 --> 00:07:04,324 But I think what's kind of interesting to me and 112 00:07:04,324 --> 00:07:09,609 this was the challenge that I found in trying to talk to my dad about it, 113 00:07:09,609 --> 00:07:15,699 was that somehow I was hurt because he he somehow felt that the important work that 114 00:07:15,699 --> 00:07:21,637 I'm doing can be broken down into the simple tasks that can just be automated. 115 00:07:21,637 --> 00:07:26,519 So somehow my hubris got a bit hurt because the type of job that 116 00:07:26,519 --> 00:07:30,660 I do is not something that a machine can replicate. 117 00:07:30,660 --> 00:07:32,212 Right? So in principle, 118 00:07:32,212 --> 00:07:36,334 I think a lot of the conversation is also stuck in this situation that 119 00:07:36,334 --> 00:07:39,299 we clearly see that machines can do quite a lot and 120 00:07:39,299 --> 00:07:44,180 they can do things faster and they can do things better and at scale and cheaper. 121 00:07:44,180 --> 00:07:46,935 But I think we're also starting to move the conversation towards. 122 00:07:46,935 --> 00:07:51,240 Yeah, but what are tasks that can be done and tasks that can't be done. 123 00:07:51,240 --> 00:07:54,745 So what are your your, do you have any any faults on that or 124 00:07:54,745 --> 00:07:57,828 is that some where do you draw the line basically? 125 00:07:57,828 --> 00:08:02,638 >> Maybe, maybe I like, I think when you say like machines can do better work, 126 00:08:02,638 --> 00:08:05,040 it's really not the point. 127 00:08:05,040 --> 00:08:09,944 I think it's exactly that that's exactly the direction if your job can be automated 128 00:08:09,944 --> 00:08:12,510 because it's made out of analytical work. 129 00:08:12,510 --> 00:08:18,210 Like if you follow a so they are checklists to fulfill a job then yes, 130 00:08:18,210 --> 00:08:21,411 Daniel Job can be automated, right? 131 00:08:21,411 --> 00:08:26,123 And then you just need people like people can enough data that can actually create 132 00:08:26,123 --> 00:08:30,913 like something that can actually make this work so cheap that then worth, Right? 133 00:08:30,913 --> 00:08:35,140 So that's I think it goes exactly the same direction of what your dad's free. 134 00:08:35,140 --> 00:08:36,728 It's already proven that directed. 135 00:08:36,728 --> 00:08:40,472 Like if you think that your job can be calculated because what are you doing is 136 00:08:40,472 --> 00:08:43,105 kind of repetitive and and can be learned quite fast, 137 00:08:43,105 --> 00:08:45,000 then you can give that to the machine. 138 00:08:45,000 --> 00:08:47,512 It doesn't mean that the machine going to do a better job than you. 139 00:08:47,512 --> 00:08:50,870 It's just gonna be that you're gonna do us cheaper job than you. 140 00:08:50,870 --> 00:08:53,461 And that's what matter promotes of the industry, 141 00:08:53,461 --> 00:08:56,482 is actually reducing the human cost for forever business. 142 00:08:56,482 --> 00:08:58,440 Right? 143 00:08:58,440 --> 00:08:59,872 >> Indeed. 144 00:08:59,872 --> 00:09:04,211 So that's that's something to consider a bit right. 145 00:09:04,211 --> 00:09:09,443 When I think I always like to go back to this image of the ladies who are taking 146 00:09:09,443 --> 00:09:14,263 the hammers towards the sewing machines back in the 19th century, 147 00:09:14,263 --> 00:09:19,020 because well, we are more or less getting into the same situation. 148 00:09:19,020 --> 00:09:23,897 And of course, I think there are a lot of people kind of cheering for 149 00:09:23,897 --> 00:09:29,556 the fact that because of that Ai can also help people maybe pick up new skills or 150 00:09:29,556 --> 00:09:34,172 maybe even move towards the society in which people have to work 151 00:09:34,172 --> 00:09:38,651 less because a lot of the technology is doing the work for us. 152 00:09:38,651 --> 00:09:43,327 So again, those are interesting perspectives to to to to consider. 153 00:09:43,327 --> 00:09:45,630 And for everyone that sees some sort of utopia, 154 00:09:45,630 --> 00:09:48,291 there's somebody else seeing some sort of dystopia. 155 00:09:48,291 --> 00:09:51,361 So I guess we will meet somewhere in the middle anyway. 156 00:09:51,361 --> 00:09:57,521 Can you maybe tell our listeners at least from your experience what is not AI. 157 00:09:57,521 --> 00:10:03,283 So we kind of explain a bit what Ai is, but what is maybe some misconception about 158 00:10:03,283 --> 00:10:08,787 AI that you maybe see a lot in in your work and then maybe you would like to say, 159 00:10:08,787 --> 00:10:13,181 hey, I think that's not really what should be put under AI. 160 00:10:13,181 --> 00:10:13,840 >> Yeah. 161 00:10:13,840 --> 00:10:18,601 I think, I mean I've definitely had clients kind of our customers come in and 162 00:10:18,601 --> 00:10:23,217 they're like, hey we want AI and I'm not necessarily saying that that's 163 00:10:23,217 --> 00:10:27,172 nothing matter, but they just want AI they know it's a thing and 164 00:10:27,172 --> 00:10:31,743 they're just like everyone like have you got a problem you want to solve? 165 00:10:31,743 --> 00:10:32,750 Have you got data? 166 00:10:32,750 --> 00:10:35,722 Like or is it just a case that you wanna be able to go and say, 167 00:10:35,722 --> 00:10:37,074 hey we do a out here today. 168 00:10:37,074 --> 00:10:38,901 Like what exactly going on. 169 00:10:38,901 --> 00:10:42,730 So I think that misconception of I think it goes back to the whole day, 170 00:10:42,730 --> 00:10:43,928 I can do everything. 171 00:10:43,928 --> 00:10:45,560 I'll solve all the problems. 172 00:10:45,560 --> 00:10:49,305 But without ever actually taking the time to figure out what your problem is. 173 00:10:49,305 --> 00:10:53,265 Is a real kind of an issue in the industry that I keep saying anyway, especially 174 00:10:53,265 --> 00:10:57,225 because what leads to that is when you take that minute to step back and actually 175 00:10:57,225 --> 00:11:01,305 figure out what your problem is, what you slowly and very quickly realize is AI is 176 00:11:01,305 --> 00:11:05,448 actually not the solution because it's a very simple problem that can be solved. 177 00:11:05,448 --> 00:11:08,812 With just the normal use cases of like whether that's development or 178 00:11:08,812 --> 00:11:11,660 like just another application or whatever else it can be. 179 00:11:11,660 --> 00:11:16,141 It's a very simple solution out there that you don't need to spend this much 180 00:11:16,141 --> 00:11:20,415 amount of time setting up an AI solution or creating these algorithms and 181 00:11:20,415 --> 00:11:21,904 it also then leads into. 182 00:11:21,904 --> 00:11:26,107 It's now becoming a scenario in the world where creating a new algorithm, 183 00:11:26,107 --> 00:11:30,380 we're creating a new machine learning thing is you're kind of reinventing 184 00:11:30,380 --> 00:11:32,736 the wheel a little bit with these things. 185 00:11:32,736 --> 00:11:35,910 If you're starting from scratch there is absolutely no need to do that. 186 00:11:35,910 --> 00:11:40,478 There are great models that are already existing like google and as your database, 187 00:11:40,478 --> 00:11:43,023 they've already got these great platforms or 188 00:11:43,023 --> 00:11:46,761 you can just kind of go in plug your data in very securely and safely. 189 00:11:46,761 --> 00:11:47,880 I promise. 190 00:11:47,880 --> 00:11:51,444 I mean I've done it is used in but it's already been built for you and yet people 191 00:11:51,444 --> 00:11:54,965 are still like, no, we want from scratch like we want to build our own thing. 192 00:11:54,965 --> 00:11:58,762 There's just so much more involved with that because of price and cost and 193 00:11:58,762 --> 00:12:02,560 all these things but also dangerous in terms of the ethics involved in like 194 00:12:02,560 --> 00:12:05,272 are you really using the best people to create this. 195 00:12:05,272 --> 00:12:07,900 Do they know the downfalls of the pitfalls here. 196 00:12:07,900 --> 00:12:11,623 So I mean there's just a lot of things that can go wrong, but also yeah, 197 00:12:11,623 --> 00:12:14,414 what is not AI is very much look at the problem at hand, 198 00:12:14,414 --> 00:12:18,420 there's probably a better solution than using a brand new AI thing into us. 199 00:12:18,420 --> 00:12:22,301 Especially when the scenario a lot of the times is a case of if not elsewhere, 200 00:12:22,301 --> 00:12:26,061 it's like if it's this, I want to answer doing this and then I want this, 201 00:12:26,061 --> 00:12:30,634 that's not AI and that's just programming, which has been around for hundreds years. 202 00:12:30,634 --> 00:12:35,216 >> Yeah, like that, but chatbots 203 00:12:35,216 --> 00:12:40,141 not AI just I got that a lot as well. 204 00:12:40,141 --> 00:12:40,842 >> All right. 205 00:12:40,842 --> 00:12:48,845 >> So what I so we kind of talked a bit about what AI is and what Ai is not? 206 00:12:48,845 --> 00:12:52,767 And then I would like to each of you because I think Akasha you kind of did 207 00:12:52,767 --> 00:12:57,354 that now when you were kind of describing the idea that okay, you have a problem and 208 00:12:57,354 --> 00:13:01,607 then you try to find a solution and AI is sort of a way to get from from problem to 209 00:13:01,607 --> 00:13:05,495 solution, but it's not necessarily the solution in and of itself. 210 00:13:05,495 --> 00:13:08,982 So I was wondering can you kind of explain a bit. 211 00:13:08,982 --> 00:13:14,400 So so for each of you kind of how how does this AI fit in what you do? 212 00:13:14,400 --> 00:13:18,696 So how can you basically how do you when you have to solve problems with AI, 213 00:13:18,696 --> 00:13:20,635 what is it that you actually do and 214 00:13:20,635 --> 00:13:24,654 what is it that you actually recommend to the to the to the companies or 215 00:13:24,654 --> 00:13:27,943 to the policymakers that come and want to work with you. 216 00:13:27,943 --> 00:13:35,515 So in this case what would be an actual application of AI that you can share? 217 00:13:35,515 --> 00:13:36,598 >> Yeah like I mean and and 218 00:13:36,598 --> 00:13:40,189 I guess I mean my experience of it is gonna be completely different because I 219 00:13:40,189 --> 00:13:44,011 work in a consulting company with specific kind of customers and stuff I guess. 220 00:13:44,011 --> 00:13:46,601 But like if I mean I call myself a data scientist but 221 00:13:46,601 --> 00:13:49,079 I mean you can call yourself whatever you want. 222 00:13:49,079 --> 00:13:49,721 I'm a consultant. 223 00:13:49,721 --> 00:13:51,833 I go talk to people all the time and 224 00:13:51,833 --> 00:13:56,440 I mean any given project the same thing happens where it's AI is just such 225 00:13:56,440 --> 00:14:01,197 a minuscule part of the project itself if it is in the first person AI project 226 00:14:01,197 --> 00:14:04,970 which is a very odd number same term in fairness a I can mean so 227 00:14:04,970 --> 00:14:09,756 many hundreds different things inside us but let's just use that for now. 228 00:14:09,756 --> 00:14:11,691 Like a lot of my job is actually going and 229 00:14:11,691 --> 00:14:14,918 doing the original chance of okay one of you got what's the data, 230 00:14:14,918 --> 00:14:18,221 what's your current tech stack like where does everything said? 231 00:14:18,221 --> 00:14:20,143 How does it all connect with everything? 232 00:14:20,143 --> 00:14:24,880 And the like it's a running joke in the data science industry where the 70% of 233 00:14:24,880 --> 00:14:27,991 the data side this job is just cleaning up the data and 234 00:14:27,991 --> 00:14:32,235 it's true like nothing works well if your data is incorrect or clean and 235 00:14:32,235 --> 00:14:34,950 actually representing the truth behind it. 236 00:14:34,950 --> 00:14:40,379 That quality of data is such a big part of any solution you'll ever build. 237 00:14:40,379 --> 00:14:44,481 It's like a lot of my job is just going, it's not even going cleaning. 238 00:14:44,481 --> 00:14:47,244 The data is actually understanding the data as a consultant. 239 00:14:47,244 --> 00:14:51,241 I'm going into this random company where these people know their data really well. 240 00:14:51,241 --> 00:14:52,399 I have no idea what they're doing. 241 00:14:52,399 --> 00:14:54,348 Like I've been in healthcare, 242 00:14:54,348 --> 00:14:57,690 I've done a project in kind of just what are they called? 243 00:14:57,690 --> 00:15:00,730 But bunch of different industries. 244 00:15:00,730 --> 00:15:02,011 I don't know what any of this means. 245 00:15:02,011 --> 00:15:05,292 Like I couldn't tell you anything that happens in a hospital outside of 246 00:15:05,292 --> 00:15:07,915 what mom as a nurse tells me that happens in the hospital or 247 00:15:07,915 --> 00:15:09,721 what I'm watching Grey's Anatomy but 248 00:15:09,721 --> 00:15:13,410 that's not enough to understand these terms put in front of me in terms of data. 249 00:15:13,410 --> 00:15:16,282 A lot of my jobs essentially because I'm a consultant, 250 00:15:16,282 --> 00:15:19,643 it's a little bit different and my job is essentially going in and 251 00:15:19,643 --> 00:15:23,798 understanding what's happening for them and after we've broken that all day, 252 00:15:23,798 --> 00:15:27,740 I was like okay this is the data, it's actually clean, it's well usable. 253 00:15:27,740 --> 00:15:30,881 It's quality data where represents the tooth and 254 00:15:30,881 --> 00:15:33,440 90% of the time that isn't the case. 255 00:15:33,440 --> 00:15:36,930 So if we get to that ideal scenario where everything lines up, 256 00:15:36,930 --> 00:15:38,481 then it's like okay cool. 257 00:15:38,481 --> 00:15:43,045 Now we can implement some kind of I mean whether it's time series analysis or 258 00:15:43,045 --> 00:15:47,822 predicting whatever else you want to do with it, that's a very small part there 259 00:15:47,822 --> 00:15:52,848 that we've done in a everything around it has been okay, let's understand this. 260 00:15:52,848 --> 00:15:55,550 And once you've implemented the AI, the rest of it is helping the client 261 00:15:55,550 --> 00:15:58,343 understand us, making sure they understand what's been implemented and 262 00:15:58,343 --> 00:15:59,700 how to use it and go forward with it. 263 00:15:59,700 --> 00:16:03,795 So in my job anyway I think yes, it's a very AI based job but 264 00:16:03,795 --> 00:16:09,411 the work I do isn't really that heavily focused necessarily all the time on AI. 265 00:16:09,411 --> 00:16:12,441 >> All right, so that was the data perspective. 266 00:16:12,441 --> 00:16:18,375 So what about the design perspective, how does your typical work day look like more? 267 00:16:18,375 --> 00:16:23,410 >> Yeah, so it really depends on which project I'm working right now. 268 00:16:23,410 --> 00:16:26,348 Like I used to be a consultant as well and 269 00:16:26,348 --> 00:16:31,830 that was when I had my first contact with AI powered application. 270 00:16:31,830 --> 00:16:34,780 Was it a pharma company in Deutschland, and 271 00:16:34,780 --> 00:16:39,190 they want to do a a medical AI as anyone else wants to do medical AI. 272 00:16:39,190 --> 00:16:44,117 But it was very interesting because the proposal there 273 00:16:44,117 --> 00:16:47,670 was we have a set of people specialist. 274 00:16:47,670 --> 00:16:51,360 So it's not like identifying traffic lights or 275 00:16:51,360 --> 00:16:55,990 under it's not kind of data that anyone can label, right? 276 00:16:55,990 --> 00:17:01,814 It was a really specific like looking at patients history and 277 00:17:01,814 --> 00:17:05,125 the patient's health history and 278 00:17:05,125 --> 00:17:11,191 then identify if a drug is causing a side effect or not, right? 279 00:17:11,191 --> 00:17:13,011 So that was the baseline. 280 00:17:13,011 --> 00:17:19,148 So you look at all the evidence that you have from the documentation and 281 00:17:19,148 --> 00:17:23,731 then you judge if what's happening there, right? 282 00:17:23,731 --> 00:17:24,640 So that's the level. 283 00:17:24,640 --> 00:17:30,392 So we have 50, only 50 specialists that has the job to train that model and 284 00:17:30,392 --> 00:17:34,816 have to at the same time, the model has to learn from that and 285 00:17:34,816 --> 00:17:37,311 have to improve itself, right? 286 00:17:37,311 --> 00:17:39,530 So that's kind of like, that was the baseline. 287 00:17:39,530 --> 00:17:43,040 But that doesn't exist in the vacuum, right? 288 00:17:43,040 --> 00:17:47,935 So that kind of the doctor or the artificial doctor or 289 00:17:47,935 --> 00:17:50,611 doesn't exist in a vacuum. 290 00:17:50,611 --> 00:17:54,694 So it has to be trained somehow. 291 00:17:54,694 --> 00:18:00,220 My job as a designer is to come in, create a space where that that specific model 292 00:18:00,220 --> 00:18:05,860 can be trained and has to get input from the doctors like the actual specialist. 293 00:18:05,860 --> 00:18:10,055 So these people has to label the data in a way that they have to label the data base 294 00:18:10,055 --> 00:18:13,042 but they have to judge if the machine is doing a good job or 295 00:18:13,042 --> 00:18:17,130 not based on the recommendation that the machine is giving, right? 296 00:18:17,130 --> 00:18:21,930 And they have to kind of define what the machine has to do. 297 00:18:21,930 --> 00:18:26,706 So I have to take that application, put that in a space where the doctors actually 298 00:18:26,706 --> 00:18:30,920 can wants to use that because no doctor going to do like labeling work for 299 00:18:30,920 --> 00:18:33,541 free and they are only 50 of those people. 300 00:18:33,541 --> 00:18:36,120 So they have to integrate that in the work full that they use. 301 00:18:36,120 --> 00:18:38,351 So basically I'm designing for two users. 302 00:18:38,351 --> 00:18:40,430 I'm designed for the doctor itself. 303 00:18:40,430 --> 00:18:42,930 So the doctor can accomplish their work. 304 00:18:42,930 --> 00:18:44,150 It's useful for them. 305 00:18:44,150 --> 00:18:45,580 It's user friendly. 306 00:18:45,580 --> 00:18:47,260 It's worth daytime. 307 00:18:47,260 --> 00:18:53,781 So it gives them a value at the same time it can train the model, right? 308 00:18:53,781 --> 00:18:59,824 So the model can profit of whatever input the doctors are giving their. 309 00:18:59,824 --> 00:19:01,192 So they can explain the reasoning that they're making decisions, 310 00:19:01,192 --> 00:19:02,130 can explain the feedback. 311 00:19:02,130 --> 00:19:07,730 It's the language that the doctor talks to the machine is understood by the machine. 312 00:19:07,730 --> 00:19:11,059 So like these are all the things that I have to take in consideration when I'm 313 00:19:11,059 --> 00:19:11,650 designing. 314 00:19:11,650 --> 00:19:16,301 So basically instead of only designing a software that would 315 00:19:16,301 --> 00:19:19,125 help the doctors to get the job done, 316 00:19:19,125 --> 00:19:24,791 I have to do something that at the same time creates value for the machine. 317 00:19:24,791 --> 00:19:27,110 So the machine also get their job done. 318 00:19:27,110 --> 00:19:29,091 So now I have two users to take considerations. 319 00:19:29,091 --> 00:19:32,931 I have to understand what the machine is also like, how the machine talks what 320 00:19:32,931 --> 00:19:37,630 the machine needs and how the machine can give back value for today user. 321 00:19:37,630 --> 00:19:42,227 So that's pretty much like a scope of a project and my daily job how does it look 322 00:19:42,227 --> 00:19:46,824 like is a lot of work in the sense of talking to the data scientist for example, 323 00:19:46,824 --> 00:19:51,245 have to sit with them the whole time to understand how the data is labeled. 324 00:19:51,245 --> 00:19:57,230 And to understand what's happening inside of the artificial brain. 325 00:19:57,230 --> 00:20:02,330 And at the same time I have to lead like the whole engineer team because actually, 326 00:20:02,330 --> 00:20:06,830 the machine learning problem is just a small slice instead of a big chunk 327 00:20:06,830 --> 00:20:09,155 of engineer problem that has there, 328 00:20:09,155 --> 00:20:12,990 because you have to have a product the end of the day, right? 329 00:20:12,990 --> 00:20:19,021 So it's like you cannot like, the product is l would say is 90% of the work in, 330 00:20:19,021 --> 00:20:25,330 10% it's just the powered by machine learning. 331 00:20:25,330 --> 00:20:30,259 >> So I think if I try to see the connections also between what you 332 00:20:30,259 --> 00:20:31,331 are saying. 333 00:20:31,331 --> 00:20:34,722 I think clearly we we see that there is a data element and 334 00:20:34,722 --> 00:20:39,518 I think both of you have have clearly specified that whatever you're doing is 335 00:20:39,518 --> 00:20:42,491 only as good as the data that you fit in, right? 336 00:20:42,491 --> 00:20:46,417 So if you don't have data that is related to the problem then or 337 00:20:46,417 --> 00:20:51,023 that is also properly classified for whatever problem you're trying to 338 00:20:51,023 --> 00:20:54,948 solve then it doesn't really matter what AI or whatever fancy 339 00:20:54,948 --> 00:20:59,571 stuff you're going to use on it because that's kind of the foundation. 340 00:20:59,571 --> 00:21:01,741 But that's only the foundation because data in and of itself is nothing. 341 00:21:01,741 --> 00:21:03,591 You also have to make sense of the data. 342 00:21:03,591 --> 00:21:07,013 So if I also from my experience with with AI projects, 343 00:21:07,013 --> 00:21:11,503 this idea of classification or labeling becomes extremely important 344 00:21:11,503 --> 00:21:16,461 because someone needs to make sense of what the data means and how do you know? 345 00:21:16,461 --> 00:21:20,588 In some cases when the data is numerical, it's relatively straightforward and 346 00:21:20,588 --> 00:21:24,130 you can kind of say, okay, this is what having this number implies. 347 00:21:24,130 --> 00:21:28,173 But in this situation where you have to go through unstructured data or 348 00:21:28,173 --> 00:21:31,751 when you have to go through data that is not so black and white. 349 00:21:31,751 --> 00:21:34,380 So this classification decision is not so simple. 350 00:21:34,380 --> 00:21:39,681 It becomes complicated and that's why when you need some sort of specialist or 351 00:21:39,681 --> 00:21:42,573 you need someone that can label the data and 352 00:21:42,573 --> 00:21:47,652 I think I find it very funny every time I have to do a capture on the internet. 353 00:21:47,652 --> 00:21:50,358 I kind of know that I'm training AI, right? 354 00:21:50,358 --> 00:21:52,921 Because it asks me where do you see a traffic light or where do you see a tree? 355 00:21:52,921 --> 00:21:56,124 When do you know and I'm clicking on all those images and 356 00:21:56,124 --> 00:22:00,571 I know that I'm kind of doing my duty to train AI to better recognize objects. 357 00:22:00,571 --> 00:22:04,883 But that's the thing, that's the labeling part that of course AI can attempt 358 00:22:04,883 --> 00:22:08,800 to do based on whatever data and whatever training said you gave it, but 359 00:22:08,800 --> 00:22:12,741 still someone needs to come back and see, okay, have you classified? 360 00:22:12,741 --> 00:22:14,351 Well how is it going? 361 00:22:14,351 --> 00:22:18,451 And in the end again, how is this whole thing helping us solve the problem? 362 00:22:18,451 --> 00:22:23,830 So it's a bit more complicated than many people would like to think. 363 00:22:23,830 --> 00:22:29,471 I also wanted to ask you, so again, you've shared experience about your work and 364 00:22:29,471 --> 00:22:33,340 and kind of what you do and the challenges that you face. 365 00:22:33,340 --> 00:22:36,081 What are some recommendations that you would give? 366 00:22:36,081 --> 00:22:41,575 Because part of our audience being at the business school are students and 367 00:22:41,575 --> 00:22:45,385 we see that many students are extremely excited and 368 00:22:45,385 --> 00:22:48,317 interested in learning more about AI. 369 00:22:48,317 --> 00:22:51,854 What would be the recommendations that you would give for 370 00:22:51,854 --> 00:22:56,054 someone who is just kind of fresh through their bachelor studies and 371 00:22:56,054 --> 00:23:00,345 they're looking forward to getting some skills into the area of AI. 372 00:23:00,345 --> 00:23:04,629 What would you recommend them to do or what what would be some resources that 373 00:23:04,629 --> 00:23:12,473 they could tap into if they want to learn more and understand more? 374 00:23:12,473 --> 00:23:15,649 >> Yeah, I think, I mean I kind of look back to when I started I guess. 375 00:23:15,649 --> 00:23:19,063 I mean, I started essentially with no tech backgrounds and 376 00:23:19,063 --> 00:23:22,067 it was really interesting because I mean I came in and 377 00:23:22,067 --> 00:23:25,510 I was basically taught everything I did on the job as I went. 378 00:23:25,510 --> 00:23:29,141 And the first thing I remember looking at in terms of specifically AI, 379 00:23:29,141 --> 00:23:33,075 before I got to the AI preminary, I mean I was obviously the only had actually 380 00:23:33,075 --> 00:23:35,597 code and learn different languages and stuff but 381 00:23:35,597 --> 00:23:38,515 for someone especially in a business context, right? 382 00:23:38,515 --> 00:23:41,198 All you really need to understand is okay, the very high levels of, 383 00:23:41,198 --> 00:23:43,851 okay, this is a computer vision model, what is going to do is XYZ. 384 00:23:43,851 --> 00:23:48,466 Which is identify images or identify objects and images and what you really, 385 00:23:48,466 --> 00:23:52,869 all you need is just a very grasp basic understanding of kind of how has that 386 00:23:52,869 --> 00:23:53,530 done it. 387 00:23:53,530 --> 00:23:58,780 You don't need to know okay, if I hyper tune this parameter in this specific code, 388 00:23:58,780 --> 00:24:02,236 this is one little tweak to actually make this happen. 389 00:24:02,236 --> 00:24:04,471 You don't need to know that, I don't even need to know that as much anymore 390 00:24:04,471 --> 00:24:06,510 because I'm not building those kind of models out on a daily basis. 391 00:24:06,510 --> 00:24:11,261 So fast AI which is a course that was built by Jeremy Howard out in Stanford I 392 00:24:11,261 --> 00:24:16,473 think, maybe I'm getting the college wrong, but he's built out this really, 393 00:24:16,473 --> 00:24:21,533 really cool course and what that has done, and I'm happy to share the links and 394 00:24:21,533 --> 00:24:25,785 stuff afterwards, but if you literally Google it pops right up. 395 00:24:25,785 --> 00:24:27,611 And it's very broken down into understandable bit. 396 00:24:27,611 --> 00:24:30,286 So I think it's like two hour lectures over seven lectures or 397 00:24:30,286 --> 00:24:31,300 something like that. 398 00:24:31,300 --> 00:24:34,031 And he essentially goes through, what does this all mean? 399 00:24:34,031 --> 00:24:35,241 Like what is computer vision? 400 00:24:35,241 --> 00:24:37,516 What is natural language processing? 401 00:24:37,516 --> 00:24:40,690 All these things in a very non coding format. 402 00:24:40,690 --> 00:24:44,070 There's a bit of coding there where you can actually tell you how it happens, but 403 00:24:44,070 --> 00:24:46,921 he goes to essentially how does it work, what's the max behind it? 404 00:24:46,921 --> 00:24:48,910 Without going to in depth into it. 405 00:24:48,910 --> 00:24:53,804 And that will be more than enough I think, to have a workable knowledge of 406 00:24:53,804 --> 00:24:58,232 what's going on in these projects from a business perspective. 407 00:24:58,232 --> 00:25:00,923 If you really want to go get really in depth into it, 408 00:25:00,923 --> 00:25:03,865 start actually doing the codes are there in the fast AI, 409 00:25:03,865 --> 00:25:08,185 start using kind of AWS Google Azure, they've all got platforms that are readily 410 00:25:08,185 --> 00:25:10,720 available to create your own models and stuff. 411 00:25:10,720 --> 00:25:14,604 Off you go, give it a shot, each of their own websites have loads of demos and 412 00:25:14,604 --> 00:25:18,734 stuff, literally, they have like step by step double so you can follow along and 413 00:25:18,734 --> 00:25:22,620 they'll give you free credits to go do all the stuff they want you to use their 414 00:25:22,620 --> 00:25:23,330 platforms. 415 00:25:23,330 --> 00:25:25,641 So there's loads of resources available. 416 00:25:25,641 --> 00:25:26,451 I think a good place. 417 00:25:26,451 --> 00:25:29,623 The 100% is fast AI, just because it really breaks it down to 418 00:25:29,623 --> 00:25:33,681 such understandable levels that once you kind of have that grasp understanding, 419 00:25:33,681 --> 00:25:36,384 they're like, okay, I can figure everything out now 420 00:25:36,384 --> 00:25:40,630 moving forward whether it's going further into it or even more abstract out of it. 421 00:25:40,630 --> 00:25:41,619 >> All right, Marah. 422 00:25:41,619 --> 00:25:47,133 >> I think it's really dependent of what where 423 00:25:47,133 --> 00:25:56,230 which part they want to take like as a fresh business person. 424 00:25:56,230 --> 00:26:00,881 I think there is one- you can think about like the applications, right? 425 00:26:00,881 --> 00:26:07,680 Like the first thing to understand is that AI is not a future, right? 426 00:26:07,680 --> 00:26:12,130 It's just one more part of the product that you're working with or working on. 427 00:26:12,130 --> 00:26:13,945 And from that from that point of view, 428 00:26:13,945 --> 00:26:17,780 like first you have to understand what kind of problem you're trying to solve. 429 00:26:17,780 --> 00:26:22,049 Like if you're trying to solve a problem then you're going the product area and 430 00:26:22,049 --> 00:26:26,127 then you have to have a understanding of, actually understanding of design 431 00:26:26,127 --> 00:26:30,931 understanding of like people's needs and what you're trying to enable there, right? 432 00:26:30,931 --> 00:26:34,111 And then you can go like more B to B that you're like working for infrastructure. 433 00:26:34,111 --> 00:26:38,230 So what kind of capacity you're trying to pry into power? 434 00:26:38,230 --> 00:26:42,887 Like let's say if you're trying to power image recognition because with that you 435 00:26:42,887 --> 00:26:45,990 can do like, you don't care about the application but 436 00:26:45,990 --> 00:26:50,000 you're trying to power image recognition as it is, as a technology. 437 00:26:50,000 --> 00:26:53,523 Then you are more working like on this backstage of the of the thing and 438 00:26:53,523 --> 00:26:55,310 what's the valued thing as well. 439 00:26:55,310 --> 00:26:57,911 So like you have to understand like where you want to be and and 440 00:26:57,911 --> 00:27:00,230 then send all of those layers first, right? 441 00:27:00,230 --> 00:27:04,598 And and then there is like plenty of resource from that point of view on online 442 00:27:04,598 --> 00:27:08,650 right now because there is a lot of people doing doing quite good work. 443 00:27:08,650 --> 00:27:13,913 There is the one that I can recommend and there's a bit biased because 444 00:27:13,913 --> 00:27:19,447 it's from Google, it's called People + AI, I can send you the link but 445 00:27:19,447 --> 00:27:24,918 people + AI research, so that the link will pair.withgoogle.com. 446 00:27:24,918 --> 00:27:29,844 And there you have like a lot of really nice guidelines of how to 447 00:27:29,844 --> 00:27:32,686 help take AI in your products like, 448 00:27:32,686 --> 00:27:37,921 from the ethical point of view from like user needs point of view. 449 00:27:37,921 --> 00:27:42,333 And I think that's super relevant if you're not going to research like, 450 00:27:42,333 --> 00:27:46,614 if you're not going into academic research on machine learning so on. 451 00:27:46,614 --> 00:27:50,220 Those things that are extremely relevant because they are your main outlet. 452 00:27:50,220 --> 00:27:53,530 Either if you're working like B to B or B to C. 453 00:27:53,530 --> 00:27:58,489 It's gonna give you the first insides of what you have to consider and 454 00:27:58,489 --> 00:28:02,861 also like understand again that AI is not a teacher, right? 455 00:28:02,861 --> 00:28:06,987 It's not something that you put that you that exists on the vacuum and 456 00:28:06,987 --> 00:28:08,360 people want to buy it. 457 00:28:08,360 --> 00:28:11,790 People actually don't care who's delivering the service, right? 458 00:28:11,790 --> 00:28:14,891 People care that the service is delivered and how it's delivered. 459 00:28:14,891 --> 00:28:19,530 Like if it's cheaper, it's like what AI does make it accessible, right? 460 00:28:19,530 --> 00:28:25,791 So, and that's the only leverage that's important for reaching markets, right? 461 00:28:25,791 --> 00:28:31,225 Like for making your product like successful is actually you don't 462 00:28:31,225 --> 00:28:37,533 deliver a better product, you just deliver it cheaper and the standardized. 463 00:28:37,533 --> 00:28:40,499 So you can set about of quality and enable and empower, 464 00:28:40,499 --> 00:28:42,710 like a lot of people to do stuff, right? 465 00:28:42,710 --> 00:28:47,550 And that's the only leverage the relying on the full AI thing, 466 00:28:47,550 --> 00:28:49,295 I think is that there. 467 00:28:49,295 --> 00:28:52,390 >> So kind of really the if I again try to summarise across what you are saying, 468 00:28:52,390 --> 00:28:55,011 it's part of the same saying but different approach, right? 469 00:28:55,011 --> 00:28:57,351 So of course, technology is fantastic and 470 00:28:57,351 --> 00:29:01,641 it's great to be aware of all the options out there and of course it depends very 471 00:29:01,641 --> 00:29:05,996 much how deep you want to get into the tech part of whether it's the coding part, 472 00:29:05,996 --> 00:29:09,340 or whether it's the hardware part, etcetera, etcetera. 473 00:29:09,340 --> 00:29:13,465 But still at the end of the day it's about solving some sort of problem and 474 00:29:13,465 --> 00:29:17,656 providing some sort of value to different stakeholders, which is more or 475 00:29:17,656 --> 00:29:21,237 less the point, or at least in my view, the point of business or 476 00:29:21,237 --> 00:29:22,821 at least of good business. 477 00:29:22,821 --> 00:29:28,699 And of course, while doing that, trying to take care of society on the whole and 478 00:29:28,699 --> 00:29:34,161 not being too heavy on the resources, so really no big challenges there. 479 00:29:34,161 --> 00:29:37,537 But if we can maybe look a bit to the future and 480 00:29:37,537 --> 00:29:42,646 you guys know quite well what AI is at the moment and what it can do and 481 00:29:42,646 --> 00:29:51,030 you probably have some very good ideas of applications that are currently running. 482 00:29:51,030 --> 00:29:56,512 Because I think that's also part of the discussion around AI. 483 00:29:56,512 --> 00:29:57,530 Right? 484 00:29:57,530 --> 00:30:02,752 If we look to 10, 20, 50 years in the future, 485 00:30:02,752 --> 00:30:07,383 what kind of problems and especially these big 486 00:30:07,383 --> 00:30:12,031 world problems if you will can AI help us with. 487 00:30:12,031 --> 00:30:17,620 So here I'm just kind of inviting you to think also maybe outside of your the scope 488 00:30:17,620 --> 00:30:23,042 of your jobs and dream in terms of, okay, you've seen what it can do, but what 489 00:30:23,042 --> 00:30:28,991 would be something really cool that we can look forward to potentially in the future. 490 00:30:28,991 --> 00:30:33,549 And here we can also think about these types of big pictures related to 491 00:30:33,549 --> 00:30:38,680 sustainability or corporate responsibility, resources, ethics, etc. 492 00:30:38,680 --> 00:30:43,811 So this is the point of the discussion where you are free to go into Sci 493 00:30:43,811 --> 00:30:48,852 fi territory and I'm very happy to hear at least what are some ideas 494 00:30:48,852 --> 00:30:54,370 that you find inspiring even though maybe they might not go to fruition. 495 00:30:54,370 --> 00:31:00,098 >> I think, the learning space like when I think about like cognitive 496 00:31:00,098 --> 00:31:07,030 work that is not scalable is education, Education is not scalable at all. 497 00:31:07,030 --> 00:31:11,274 You can be exposed to content and then you learn something but there is no 498 00:31:11,274 --> 00:31:16,100 one there to tell you if you're learning something or not until you practice it. 499 00:31:16,100 --> 00:31:17,030 Right? 500 00:31:17,030 --> 00:31:20,774 And I think making education more accessible so 501 00:31:20,774 --> 00:31:24,892 that more people in the global files for example, or 502 00:31:24,892 --> 00:31:29,947 I don't know someone in the countryside in brazil and brazil and 503 00:31:29,947 --> 00:31:35,470 by the way, a person countryside in brazil can learn whatever skill or 504 00:31:35,470 --> 00:31:42,021 can learn whatever profession they want to learn and having like this a campaign and 505 00:31:42,021 --> 00:31:48,597 like engage experience that you would have with a tutor I think it's mind blowing. 506 00:31:48,597 --> 00:31:50,203 Right? Or for example, 507 00:31:50,203 --> 00:31:54,051 also learning a language that doesn't exist anymore. 508 00:31:54,051 --> 00:31:57,263 Right? Having the chance of experiencing culture 509 00:31:57,263 --> 00:32:02,422 or experience a content that's not there anymore or it's like only a few people 510 00:32:02,422 --> 00:32:07,896 know about that and you have the chance to practice that from whatever place you are. 511 00:32:07,896 --> 00:32:13,690 That is so wholesome and I kind of hate you for saying that because I'm like. 512 00:32:13,690 --> 00:32:14,593 [INAUDIBLE]. 513 00:32:14,593 --> 00:32:18,590 I'm sitting here, I'm like, what is going to happen? 514 00:32:18,590 --> 00:32:20,664 And I think, I mean I say this all time, I'm like, 515 00:32:20,664 --> 00:32:23,656 I don't know what my own five year future is going to look like because I can 516 00:32:23,656 --> 00:32:25,930 barely tell you what it's going to happen tomorrow. 517 00:32:25,930 --> 00:32:30,027 Everything just keeps changing on me so I couldn't even tell you, like three years 518 00:32:30,027 --> 00:32:33,970 ago, I never saw myself being in tech, let alone talking about AI and Podcast. 519 00:32:33,970 --> 00:32:39,266 But I think I'm a big passionate person about talking around auto ML and 520 00:32:39,266 --> 00:32:44,035 low code applications and kind of essentially democratizing AI 521 00:32:44,035 --> 00:32:48,888 like that is a very passion project for me and 90% of the talks, 522 00:32:48,888 --> 00:32:53,250 the tech talks I tend to do surround that aspect of things. 523 00:32:53,250 --> 00:32:58,689 So yeah, I mean what that is essentially enabling is, look, there's data everywhere 524 00:32:58,689 --> 00:33:03,610 I produce data myself about myself on a daily basis that I can't even leverage, 525 00:33:03,610 --> 00:33:08,091 like as a day to sign doesn't mean I can download all my chat on Facebook or 526 00:33:08,091 --> 00:33:12,719 whatever else and I don't even know what I can do with it because I'm like, 527 00:33:12,719 --> 00:33:16,905 it's there, it's gonna take me years and like computer powers and 528 00:33:16,905 --> 00:33:20,620 going on doing all this kind of stuff to do anything with it. 529 00:33:20,620 --> 00:33:25,184 And the hope obviously is like the more into the future we get the fact that there 530 00:33:25,184 --> 00:33:29,956 is the Azure, I keep saying Azure I'm just so used to saying Azure but like cloud, 531 00:33:29,956 --> 00:33:34,449 is available to you to not have to buy a big hefty machine to not be able to like, 532 00:33:34,449 --> 00:33:38,944 I mean it's very quick to loan that out and just use it to whatever you need it to 533 00:33:38,944 --> 00:33:43,651 be, whether that's machine learning or whether that's just surfing the web, 534 00:33:43,651 --> 00:33:47,870 like whatever else it might be, it's reducing that barrier to entry. 535 00:33:47,870 --> 00:33:52,443 And then once you're in there, I mean I've seen people who are doctors who studied 536 00:33:52,443 --> 00:33:56,495 their whole lives into being doctors, they're experts and say cancer or 537 00:33:56,495 --> 00:33:58,530 whatever else that might be. 538 00:33:58,530 --> 00:34:01,943 They don't need to then be like connected to someone who is a data scientist, 539 00:34:01,943 --> 00:34:05,303 if we make these things really accessible and low code, they can actually go 540 00:34:05,303 --> 00:34:08,979 leverage their own machine learning models to help them identify specific cells or 541 00:34:08,979 --> 00:34:12,131 whatever else it might be it's just getting that power to everyone else 542 00:34:12,131 --> 00:34:14,690 like really democratizing it I think it's really cool. 543 00:34:14,690 --> 00:34:19,221 You don't need to have spent whatever many years being a PhD in machine learning or 544 00:34:19,221 --> 00:34:23,162 math or whatever else to really understand and use that kind of stuff for 545 00:34:23,162 --> 00:34:27,626 yourself it's gonna be something really cool I think there's a lot of work being 546 00:34:27,626 --> 00:34:30,712 done in that space to make it as easy as possible to use and 547 00:34:30,712 --> 00:34:34,127 leverage that thing I hope that happens, but if it doesn't, 548 00:34:34,127 --> 00:34:37,190 I'll keep harping about it and I'll die on that hill. 549 00:34:37,190 --> 00:34:38,530 >> Okay. 550 00:34:38,530 --> 00:34:41,870 Yeah, and I would like that to build on that like, 551 00:34:41,870 --> 00:34:46,740 this democratizing has a lot to do as well with making things invisible. 552 00:34:46,740 --> 00:34:48,996 Right? Because the moment that it's so 553 00:34:48,996 --> 00:34:51,607 evident that you are using this AI thing and so 554 00:34:51,607 --> 00:34:56,370 on, it's kind of like push away people, if people don't want to deal with that. 555 00:34:56,370 --> 00:34:58,986 It's like imagine that every time you want to make a phone call, 556 00:34:58,986 --> 00:35:01,464 you have to understand about micro computing engineering. 557 00:35:01,464 --> 00:35:02,512 Right? It's like, no, 558 00:35:02,512 --> 00:35:05,807 you're never going to do the phone call so you don't achieve your goals, 559 00:35:05,807 --> 00:35:09,420 you don't connect to people, you don't do this stuff because that is too much in 560 00:35:09,420 --> 00:35:11,972 between there's too much cognitive load between you and 561 00:35:11,972 --> 00:35:13,649 the things that you want to achieve. 562 00:35:13,649 --> 00:35:15,850 Right? But the moment that the thing is there and 563 00:35:15,850 --> 00:35:19,611 is invisible and it's like how much of the things can you actually care about? 564 00:35:19,611 --> 00:35:25,498 For example going back to the doctor thing like analyzing the cells how much 565 00:35:25,498 --> 00:35:31,868 they can spare time with like dual jobs and actually spend time with the patient. 566 00:35:31,868 --> 00:35:32,629 Right? Like and 567 00:35:32,629 --> 00:35:35,633 take care of like the things that machine cannot do that's being empathic with 568 00:35:35,633 --> 00:35:38,630 a human being that's suffering right now from cancer for example. 569 00:35:38,630 --> 00:35:41,217 Right? So the whole the whole idea like of 570 00:35:41,217 --> 00:35:45,154 personality shifts because with everyone can be a doctor 571 00:35:45,154 --> 00:35:48,681 like in the sense of gathering having the skills and 572 00:35:48,681 --> 00:35:52,538 the technical knowledge to understand the diseases and 573 00:35:52,538 --> 00:35:57,900 the final diagnosed and they find the treatment when you lower that barrier. 574 00:35:57,900 --> 00:36:01,978 You enable a lot of people to develop what other kind of skills that are way more 575 00:36:01,978 --> 00:36:05,888 necessary than actually knowing if you like because it's a prime error. 576 00:36:05,888 --> 00:36:08,501 Right? The medicine like if you give them a pill 577 00:36:08,501 --> 00:36:11,289 doesn't work, give them another treatment 578 00:36:11,289 --> 00:36:15,350 I saw the study that if you give them like five pills it works you know? 579 00:36:15,350 --> 00:36:19,824 And and actually focus on carrying that the person gets better and carrying how 580 00:36:19,824 --> 00:36:24,363 they work anyway so, that I think like invisibility gonna be the future of AI and 581 00:36:24,363 --> 00:36:28,632 it's going to be like oxygen like you don't think about breathing if you go 582 00:36:28,632 --> 00:36:32,430 around it's just there and you enjoy it. 583 00:36:32,430 --> 00:36:34,919 >> I think to that point it's kind of already happening. 584 00:36:34,919 --> 00:36:37,354 Right? I mean I don't know there's not many 585 00:36:37,354 --> 00:36:42,146 people who don't necessarily look at their phone I mean to open my phone I have face 586 00:36:42,146 --> 00:36:46,400 ID locked on there and I think about it that's AI in built into our lives. 587 00:36:46,400 --> 00:36:50,549 The amount of people who will never consider that as an AI in their life is 588 00:36:50,549 --> 00:36:54,215 bizarre to me to be honest are like the homes like the smartphone 589 00:36:54,215 --> 00:36:55,621 things already there. 590 00:36:55,621 --> 00:36:59,279 People use it constantly and never think about the fact that that's natural 591 00:36:59,279 --> 00:37:03,568 language processing happening because it's listening to me and understanding that and 592 00:37:03,568 --> 00:37:07,285 converting that to other things it's just there it's always start to happen 593 00:37:07,285 --> 00:37:11,060 it would be really interesting how much it will actually implement itself into 594 00:37:11,060 --> 00:37:12,010 our lives I guess. 595 00:37:12,010 --> 00:37:13,830 >> It's also like AI. >> Yeah, that's right. 596 00:37:13,830 --> 00:37:18,324 >> OR the snapchats or Instagram futures like all of 597 00:37:18,324 --> 00:37:23,037 those stuff are they already and people enjoy it and 598 00:37:23,037 --> 00:37:29,130 it's kind of making their life joyful and youthful and so on. 599 00:37:29,130 --> 00:37:29,771 >> All right? 600 00:37:29,771 --> 00:37:34,483 So we are slowly moving towards the conclusion of our discussion and 601 00:37:34,483 --> 00:37:39,608 I think there were a lot of very interesting things that were mentioned and 602 00:37:39,608 --> 00:37:44,486 I would just like to kind of address some specific questions now to each 603 00:37:44,486 --> 00:37:47,792 of you towards towards the end because I think, 604 00:37:47,792 --> 00:37:52,502 you mentioned something which to me is something that I had to always 605 00:37:52,502 --> 00:37:57,380 learn the hard way, especially part of this research work that I do and 606 00:37:57,380 --> 00:38:02,917 also working with students where I see a lot of enthusiasm about a lot of things, 607 00:38:02,917 --> 00:38:07,900 but I find myself saying this quite a lot, don't reinvent the wheel. 608 00:38:07,900 --> 00:38:12,335 So every time I have a project with students in which we're trying to, develop 609 00:38:12,335 --> 00:38:16,838 a service that's going to help solve some sort of problem it's always about and 610 00:38:16,838 --> 00:38:20,825 now let's create this app and this app is going to change everything. 611 00:38:20,825 --> 00:38:22,568 Right? And I always say, okay, 612 00:38:22,568 --> 00:38:26,721 but have you considered all the other options that are already out there? 613 00:38:26,721 --> 00:38:28,201 Do you need to create the new app? 614 00:38:28,201 --> 00:38:32,261 Why would people use the up and so on and so forth so, when it comes to AI 615 00:38:32,261 --> 00:38:36,401 what are your recommendations in terms of not reinventing the wheel? 616 00:38:36,401 --> 00:38:40,758 I think that everybody, and this is again where hubris comes in, 617 00:38:40,758 --> 00:38:45,671 everybody wants to create the next big kind of thing that is going to change 618 00:38:45,671 --> 00:38:49,237 everything, but how to avoid that and how to really, 619 00:38:49,237 --> 00:38:53,065 make these steps towards making things better and yeah. 620 00:38:53,065 --> 00:38:56,412 >> Yeah, I mean, I think more, I would agree, like, 621 00:38:56,412 --> 00:38:59,927 I mean anything we're creating is to solve a problem. 622 00:38:59,927 --> 00:39:01,522 Right? And it's a people problem, 623 00:39:01,522 --> 00:39:04,624 like that's where the whole design stuff comes in, which I mean design and 624 00:39:04,624 --> 00:39:07,530 data work so well together and like not enough people talk about that, 625 00:39:07,530 --> 00:39:10,120 which is why I'm glad you had these two different sides of it. 626 00:39:10,120 --> 00:39:13,724 But yeah, that's the essence of it, is like you're trying to solve a problem, 627 00:39:13,724 --> 00:39:17,331 not necessarily that you're building something that is like innovatively cool 628 00:39:17,331 --> 00:39:17,990 no one cares. 629 00:39:17,990 --> 00:39:21,357 Dyson does really, really cool stuff I love their hair dryer I paid ridiculous 630 00:39:21,357 --> 00:39:24,981 amounts of money for it, but I don't sit there thinking about the technology behind 631 00:39:24,981 --> 00:39:27,720 him, just like, hey, it's doing this job by drying my hair. 632 00:39:27,720 --> 00:39:32,173 So the same kind of goes with AI nobody cares if you've spent literally three 633 00:39:32,173 --> 00:39:36,906 years building out this perfect algorithm that is 98% accurate when you could 634 00:39:36,906 --> 00:39:41,777 have literally just leverage, say the auto ml that as your offers, which gives you 635 00:39:41,777 --> 00:39:46,717 96% accuracy and have it done in two days, no one's gonna care actually they will 636 00:39:46,717 --> 00:39:51,446 care because hey, it took three years to do this, I could have had this two days, 637 00:39:51,446 --> 00:39:54,840 like two years ago that kind of thing plays into it for me. 638 00:39:54,840 --> 00:39:58,910 So the things I've noticed and I've used them in my own projects and stuff Azure 639 00:39:58,910 --> 00:40:03,102 has done really, really cool stuff I'm sure other platforms have done that too, 640 00:40:03,102 --> 00:40:07,050 but I guess this is my bias coming through is just what I used predominantly but 641 00:40:07,050 --> 00:40:10,451 auto ML by then the cognitive services like it's just click four or 642 00:40:10,451 --> 00:40:14,158 five buttons and you've got a machine learning model that's ready to be 643 00:40:14,158 --> 00:40:16,880 implemented into an app you already have existing. 644 00:40:16,880 --> 00:40:21,230 It is crazy cool stuff with power comes responsibility please don't 645 00:40:21,230 --> 00:40:24,080 forget that there is ethics involved here but 646 00:40:24,080 --> 00:40:29,111 I mean at the same time it is removing a lot of bias about having already created. 647 00:40:29,111 --> 00:40:31,641 I'm hoping the underneath machine learning models 648 00:40:31,641 --> 00:40:34,629 are not biased that's a whole different issue altogether but 649 00:40:34,629 --> 00:40:38,365 those things exist cognitive services their cognitive searches there where 650 00:40:38,365 --> 00:40:42,030 you can literally implement any kind of really detailed search. 651 00:40:42,030 --> 00:40:45,755 I think it took me 10 minutes to set one up the other day where I was creating 652 00:40:45,755 --> 00:40:49,779 a demo for something ready to go it will search through whatever document you want 653 00:40:49,779 --> 00:40:53,821 to put into it within your own context and that's all it takes and I recommend it. 654 00:40:53,821 --> 00:40:58,075 Like there's so many tutorials I've done talks that are available online for 655 00:40:58,075 --> 00:41:01,944 free go read about it like they made the effort to actually make it really 656 00:41:01,944 --> 00:41:06,328 understandable and very quickly and easily consumable as well at the same time so 657 00:41:06,328 --> 00:41:09,830 that's what I would say it's my suggestion I guess. 658 00:41:09,830 --> 00:41:10,504 >> All right? 659 00:41:10,504 --> 00:41:15,389 >> And Morrow you said something that really sparked a kind of a light bulb in 660 00:41:15,389 --> 00:41:20,593 my head, you said that when you're designing you're not just designing for 661 00:41:20,593 --> 00:41:25,110 the human element, but you're also designing for the AI client. 662 00:41:25,110 --> 00:41:29,905 And can you tell us a bit more about how is it to design for 663 00:41:29,905 --> 00:41:34,281 AI as in it's a very empathic way of looking at it, 664 00:41:34,281 --> 00:41:38,763 which I haven't seen in a lot of these tech talks and 665 00:41:38,763 --> 00:41:43,558 a lot of these articles that I'm reading about how AI is 666 00:41:43,558 --> 00:41:48,562 going to take over and do all sorts of nefarious things and 667 00:41:48,562 --> 00:41:54,131 you're talking about designing with the value for AI in mind. 668 00:41:54,131 --> 00:41:57,030 So can you just expand about that a bit? 669 00:41:57,030 --> 00:41:57,690 >> Yes, sir. 670 00:41:57,690 --> 00:42:03,098 If you have a kid, you don't understand that very well it's the same thing. 671 00:42:03,098 --> 00:42:04,443 Right? Like you said, 672 00:42:04,443 --> 00:42:08,652 you're telling your kids like if you put your finger in the power plug, 673 00:42:08,652 --> 00:42:12,526 you're gonna get a shock and they have no data evidence for that. 674 00:42:12,526 --> 00:42:15,653 Right? And then so they cannot make the judgment, 675 00:42:15,653 --> 00:42:19,415 they cannot identify where is danger, where it's not and 676 00:42:19,415 --> 00:42:24,712 that's what you do with AI it's exactly the same logic of educating these small, 677 00:42:24,712 --> 00:42:28,624 really simple minded being in to do a task that you want to do so 678 00:42:28,624 --> 00:42:33,191 different from a kid that to learn from everything in sight response. 679 00:42:33,191 --> 00:42:38,378 They I only have the data that you have there and only talk to one specific 680 00:42:38,378 --> 00:42:42,973 language and cannot understand really simple correlations. 681 00:42:42,973 --> 00:42:47,011 Right? Every time you try to add one more data 682 00:42:47,011 --> 00:42:52,124 point one more labelling it's just like everything 683 00:42:52,124 --> 00:42:57,468 that's just more messy and messy with the accuracy and 684 00:42:57,468 --> 00:43:04,569 also like it's kind of like breaks the service that you want to deliver. 685 00:43:04,569 --> 00:43:05,151 Right? 686 00:43:05,151 --> 00:43:09,855 So you're training a really dumb employee to deliver a job that's 687 00:43:09,855 --> 00:43:14,643 what you're trying to do that space so you have to be really clear and 688 00:43:14,643 --> 00:43:19,431 you have to think like what are the data entry points that the machine 689 00:43:19,431 --> 00:43:24,230 needs in order to learn in order to perform the job better. 690 00:43:24,230 --> 00:43:26,250 Right? Like how can you make that effort? 691 00:43:26,250 --> 00:43:28,839 Of course there is like a lot of deep learning and 692 00:43:28,839 --> 00:43:32,466 things that make the machine learned by itself but still like how to 693 00:43:32,466 --> 00:43:37,730 create the applications that create the interface between user machine backwards. 694 00:43:37,730 --> 00:43:42,089 Like every time you use something in google you'll give feedback. 695 00:43:42,089 --> 00:43:45,655 Right? That's the data point every time your tax, 696 00:43:45,655 --> 00:43:50,700 you write something on the google docs, you're creating a data point 697 00:43:50,700 --> 00:43:55,674 of of like how the world works or your training your spell corrector. 698 00:43:55,674 --> 00:43:56,250 Right? So 699 00:43:56,250 --> 00:44:00,767 all of those things they are data points that go into the model if the model well 700 00:44:00,767 --> 00:44:05,632 built, we're gonna learn faster and less and less, you have this idea that we're 701 00:44:05,632 --> 00:44:09,733 talking like, if it's well done the model, then you have less design 702 00:44:09,733 --> 00:44:14,630 interface the machine gets more and more invisible and that's your goal. 703 00:44:14,630 --> 00:44:19,210 And then of the day you want to have like something that the system is not seeing 704 00:44:19,210 --> 00:44:20,493 things just happen. 705 00:44:20,493 --> 00:44:21,330 Right? 706 00:44:21,330 --> 00:44:24,685 It's a bit like to think in a car drive and 707 00:44:24,685 --> 00:44:28,347 a self driving car, like if you grab a taxi. 708 00:44:28,347 --> 00:44:29,980 Right? You get a cab or 709 00:44:29,980 --> 00:44:34,408 Uber or something to go to the airport, you actually, 710 00:44:34,408 --> 00:44:38,469 most of the time you don't care about the driver. 711 00:44:38,469 --> 00:44:40,630 Right? You just want to get to your location, 712 00:44:40,630 --> 00:44:43,986 not that you like when you're ordering a cab, you go like a bit like Tinder, 713 00:44:43,986 --> 00:44:47,290 I don't like this driver I don't like to drive and now, I'm gonna get it, 714 00:44:47,290 --> 00:44:48,192 you don't do that. 715 00:44:48,192 --> 00:44:50,335 Right? Or you like to see cars passing and 716 00:44:50,335 --> 00:44:53,117 say, I like this car now, I don't like this car, 717 00:44:53,117 --> 00:44:56,384 you don't care you just want to get to where you want to go. 718 00:44:56,384 --> 00:44:58,989 Right? And that's the main thing about self 719 00:44:58,989 --> 00:45:03,821 driving cars as well as that there is no difference between self driving car and 720 00:45:03,821 --> 00:45:07,056 a taxi from the perspective of getting the job done. 721 00:45:07,056 --> 00:45:07,667 Right? And 722 00:45:07,667 --> 00:45:11,462 that's the overall goal overarching goal is like to be so 723 00:45:11,462 --> 00:45:18,130 invisible that it's indistinguishable from a service delivery from a person. 724 00:45:18,130 --> 00:45:18,846 >> All right? 725 00:45:18,846 --> 00:45:23,662 So to kind of bring everything together, whether it's from a data 726 00:45:23,662 --> 00:45:28,478 science perspective or whether it's from a design perspective, 727 00:45:28,478 --> 00:45:33,380 it's still kind of boils down to figuring out what the problem is and 728 00:45:33,380 --> 00:45:38,470 how do you solve that problem and how do you offer value in the process? 729 00:45:38,470 --> 00:45:42,670 While of course leveraging all the technology and 730 00:45:42,670 --> 00:45:47,670 the complicated machine learning and algorithm and so on, 731 00:45:47,670 --> 00:45:52,370 but while still having this human element that either is 732 00:45:52,370 --> 00:45:57,170 part of defining the problem or making sense of the data or 733 00:45:57,170 --> 00:46:01,770 then later on evaluating and providing feedback and so 734 00:46:01,770 --> 00:46:06,200 on and, determine to what extent value is created. 735 00:46:06,200 --> 00:46:10,877 So still quite a human problem in the end of human problem, 736 00:46:10,877 --> 00:46:15,184 human solution and again, trying to provide value. 737 00:46:15,184 --> 00:46:20,087 I want to thank both of you for this talk again, it's been so 738 00:46:20,087 --> 00:46:22,834 fun hearing your perspective and 739 00:46:22,834 --> 00:46:28,033 trying to kind of bridge them and I think the bridges were fairly 740 00:46:28,033 --> 00:46:32,938 clear if people want to hear more about what you're doing or 741 00:46:32,938 --> 00:46:37,971 see other talks that you've done, where can they find you? 742 00:46:37,971 --> 00:46:39,880 We will of course also have links, but 743 00:46:39,880 --> 00:46:45,630 if there's anything that you would like to plug now this would be the time. 744 00:46:45,630 --> 00:46:50,152 >> Yeah, I think twitter like to enrich out wherever I'm always happy to help, 745 00:46:50,152 --> 00:46:54,257 preferably Twitter, Linkedin gets clogged up very quickly but hey, 746 00:46:54,257 --> 00:46:58,777 I mean if you want to hear more from me, I run my own podcast as well actually and 747 00:46:58,777 --> 00:47:02,211 it's a bit more around different career paths into tech. 748 00:47:02,211 --> 00:47:06,126 So if you are out there listening and you might want to switch into a career in tech 749 00:47:06,126 --> 00:47:09,926 come along this and I'm sure you'll find somebody who's had the same kind of 750 00:47:09,926 --> 00:47:12,530 journey as you in a wild on traditional way into it. 751 00:47:12,530 --> 00:47:16,170 I mean the one that's coming out tomorrow is a ballet dancer who's turned into an AI 752 00:47:16,170 --> 00:47:17,630 person so it's very cool stuff. 753 00:47:17,630 --> 00:47:19,860 >> Feel free to also drop the name. 754 00:47:19,860 --> 00:47:21,774 So what is your podcast called? 755 00:47:21,774 --> 00:47:23,345 >> Yes. >> And where can you find it? 756 00:47:23,345 --> 00:47:25,365 >> Shamelessly plugging myself and 757 00:47:25,365 --> 00:47:29,530 not even knowing how to do it path uncovered was the podcast. 758 00:47:29,530 --> 00:47:30,935 >> All right. Path uncovered. 759 00:47:30,935 --> 00:47:31,551 Thank you. 760 00:47:31,551 --> 00:47:35,690 Morrow where can people find you and see what you're thinking and doing? 761 00:47:35,690 --> 00:47:40,796 >> I think Twitter and Instagram are the true main space where I share stuff so 762 00:47:40,796 --> 00:47:46,090 if you just find them all relax and none of those stuff you're gonna find me. 763 00:47:46,090 --> 00:47:51,700 I have some talks scheduled next month in the San Francisco design 764 00:47:51,700 --> 00:47:57,718 week we're going to talk exactly about what's AI for for designers, 765 00:47:57,718 --> 00:48:04,830 so it's gonna be like a more 45 minutes me talking about exactly this topic. 766 00:48:04,830 --> 00:48:10,505 It's going to be available line and and people can can watch it from anywhere 767 00:48:10,505 --> 00:48:16,270 besides that I have also podcast, but only I only have five episodes there and 768 00:48:16,270 --> 00:48:22,212 it's unfortunately in bad German where we talk about education technology and 769 00:48:22,212 --> 00:48:27,533 basically there is one episode that's exactly about what's the AI for 770 00:48:27,533 --> 00:48:33,430 education and how that can impact the whole life sort of schools. 771 00:48:33,430 --> 00:48:37,397 >> If someone wants to practice their bad German, what would be the podcast. 772 00:48:37,397 --> 00:48:39,090 What's the name of the podcast? 773 00:48:39,090 --> 00:48:41,830 >> It's called Buildings, hack. 774 00:48:41,830 --> 00:48:42,850 >> Buildings, hack. 775 00:48:42,850 --> 00:48:43,584 All right? 776 00:48:43,584 --> 00:48:44,180 Awesome. 777 00:48:44,180 --> 00:48:48,649 So we will also of course provide all the all the links in the description so 778 00:48:48,649 --> 00:48:52,612 I want to thank both of you again for taking the time to talk to us and 779 00:48:52,612 --> 00:48:56,649 we look forward to also hearing what some of our listeners think and 780 00:48:56,649 --> 00:48:59,360 what the conversation around that will be. 781 00:48:59,360 --> 00:49:00,971 So thanks again. 782 00:49:00,971 --> 00:49:03,980 And yeah, that's it. 783 00:49:03,980 --> 00:49:05,968 Thanks for inviting Robert. 784 00:49:05,968 --> 00:49:06,860 >> [INAUDIBLE]. 785 00:49:06,860 --> 00:49:07,551 >> Thank you. 786 00:49:07,551 --> 00:49:08,075 [INAUDIBLE]. 787 00:49:08,075 --> 00:49:28,596 [MUSIC].