WEBVTT - Google Goes Magenta 0:00:00.160 --> 0:00:07.160 Brought to you by Toyota. Let's go places. Welcome to 0:00:07.360 --> 0:00:14.920 Forward Thinking. Hey there, and welcome to Forward Thinking, the 0:00:15.120 --> 0:00:18.239 podcast that looks at the future and says rewritten by 0:00:18.239 --> 0:00:24.840 a machine with new technology. I'm John and I'm Joe McCormick. So, guys, 0:00:25.239 --> 0:00:28.560 you know, we've talked a little bit about AI and 0:00:28.840 --> 0:00:32.400 its potential role in creativity a few times. Back in September, 0:00:33.320 --> 0:00:36.400 we recorded an episode called the Future of Music Composition, 0:00:36.920 --> 0:00:39.559 and we talked about electronically aided composition as well as 0:00:39.600 --> 0:00:43.400 the idea of AI being able to compose music all 0:00:43.440 --> 0:00:45.879 by itself. You know. One of the funny things is 0:00:46.400 --> 0:00:49.159 on that episode, I think we we talked about, you know, 0:00:49.200 --> 0:00:52.599 will we ever have something that and and then it's 0:00:52.640 --> 0:00:55.600 stuff we already have right now. Yeah. Yeah, to be fair, 0:00:55.680 --> 0:00:57.880 some of it may not have been as as publicly 0:00:57.920 --> 0:01:01.400 available in as it is. No, I mean now today, 0:01:01.520 --> 0:01:07.080 not necessarily now, back then we had earlier versions of them. Yeah. Oh, 0:01:07.400 --> 0:01:09.399 the base base technology that we're going to be talking 0:01:09.440 --> 0:01:13.600 about today, really deep learning didn't kind of was not 0:01:13.680 --> 0:01:18.880 announced to the people. Yeah, we can. We cannot be blamed, 0:01:19.360 --> 0:01:22.000 is I think what we're trying to say here, But 0:01:22.120 --> 0:01:26.240 we're futurists, not psychics. We're not even really futurists. Reporters, Yeah, 0:01:26.240 --> 0:01:29.480 we're not even really reporters. Kind of sit in front 0:01:29.480 --> 0:01:32.080 of microphone. We're just gonna keep stepping back. We don't 0:01:32.080 --> 0:01:34.800 know what we are, but we know what we're excited about. 0:01:35.000 --> 0:01:38.360 We're excited about this idea of AI being able to 0:01:38.440 --> 0:01:42.920 engage in works of creativity. And before, just before we 0:01:43.040 --> 0:01:46.560 jumped into the studio, Joe, you came across an example 0:01:46.760 --> 0:01:50.560 of AI that works with music. Yeah, so I I 0:01:50.640 --> 0:01:53.560 was wondering if you guys had listened to any of 0:01:53.600 --> 0:01:57.600 the creations of this app called juke deck, And before 0:01:57.720 --> 0:02:00.640 you asked us, I had not, But Lauren and I 0:02:00.680 --> 0:02:04.960 listened to a track. I decided to generate a music 0:02:05.040 --> 0:02:09.840 track that was in the electronica genre with a mood 0:02:10.280 --> 0:02:14.080 of aggressive. Was it aggressive, No, it was kind of. 0:02:14.160 --> 0:02:18.920 It was kind of a little bouncy yeah, yeah, aggressively bouncy. Yeah. 0:02:19.120 --> 0:02:22.560 What juke Dick does is it will let you specify 0:02:22.560 --> 0:02:26.040 a mood and specify a genre from like, you know, 0:02:26.120 --> 0:02:28.800 four different choices. You can pick a piano thing, a 0:02:28.880 --> 0:02:31.720 folk thing, an electronic thing, or something else. I can't 0:02:31.720 --> 0:02:34.200 remember the last one, and then it'll let you pick 0:02:34.200 --> 0:02:35.919 a mood to go with that and say how long 0:02:35.960 --> 0:02:38.760 of a track you want, and then it'll just generate 0:02:38.960 --> 0:02:42.720 for you an original track of music that is royalty 0:02:42.800 --> 0:02:46.880 free that was created by a I. So I first 0:02:46.919 --> 0:02:49.480 told it I wanted a ninety second original track of 0:02:49.520 --> 0:02:52.840 electronic music for a chilled mood, and it gave me 0:02:52.880 --> 0:02:57.200 a track called Holistic Adventure. Ours was called Sweltering Seas. 0:02:57.520 --> 0:03:00.720 That's pretty good. And then it also made me a 0:03:00.800 --> 0:03:04.240 sixty second original track of folk style music for a 0:03:04.360 --> 0:03:08.359 melancholic mood, which was called Infinite Atoms. And beyond that, 0:03:08.600 --> 0:03:11.160 if you were to go further into it, they actually 0:03:11.240 --> 0:03:16.320 have other genres available as well, including cinematic, and they're uh, 0:03:16.880 --> 0:03:18.840 just to pay up for those. Yeah, it's it's when 0:03:18.840 --> 0:03:22.040 you've subscribed to the service. And one thing we should 0:03:22.080 --> 0:03:24.960 point out is that not every mood is available with 0:03:25.040 --> 0:03:27.720 every style of music, So you know, you can't really 0:03:27.720 --> 0:03:31.240 necessarily get a lot of aggressive folk necessarily. Yeah, So 0:03:31.320 --> 0:03:34.000 I'm not exactly sure how this works. So the way 0:03:34.040 --> 0:03:37.280 they explain it is, quote, our AI uses machine learning 0:03:37.320 --> 0:03:39.880 to understand how to write music chord by chord and 0:03:39.960 --> 0:03:42.840 note by note, this means that every track you create 0:03:42.960 --> 0:03:45.880 using juke Deck is truly unique, and so what they're 0:03:45.920 --> 0:03:50.760 saying is that it truly is ground up auto generated. Um, 0:03:50.840 --> 0:03:54.960 you know, artificial intelligence created music and they're not just 0:03:55.080 --> 0:03:59.400 like working from templates or something like that. But then again, 0:03:59.440 --> 0:04:02.120 I mean, it's it's hard to know what what's really 0:04:02.120 --> 0:04:04.360 going on on the back end, but at least if 0:04:04.400 --> 0:04:08.160 it works as advertised, I'm impressed. Neither one of the 0:04:08.160 --> 0:04:10.840 tracks I heard was like great music it you know, 0:04:10.880 --> 0:04:13.920 it wasn't blowing my mind. I wasn't just amazed, oh man, 0:04:13.960 --> 0:04:17.440 that's so good, but it was absolutely passable for its genre, 0:04:17.520 --> 0:04:20.040 totally good enough to be elevator music or the kind 0:04:20.040 --> 0:04:22.040 of music you'd playing a store or something like that. 0:04:22.080 --> 0:04:24.160 I was telling Lauren that it reminded me of the 0:04:24.200 --> 0:04:26.320 kind of music you would encounter in like a small 0:04:26.400 --> 0:04:29.560 independent video games, Like you're playing a little video game 0:04:29.560 --> 0:04:31.599 where clearly it's maybe one or two people who have 0:04:31.680 --> 0:04:34.160 worked on that, and it totally was something that you 0:04:34.160 --> 0:04:37.640 would hear from that. Not bad, not like a spectacular 0:04:37.760 --> 0:04:39.680 entry into the field, as you were saying, Joe, but 0:04:39.720 --> 0:04:43.640 not bad. Um. It's also I don't know for a 0:04:43.720 --> 0:04:45.760 fact that you can do this, But I was playing 0:04:45.760 --> 0:04:47.279 with it and it looked like there might even be 0:04:47.360 --> 0:04:50.400 the possibility of combining genres. So I wanted to find 0:04:50.400 --> 0:04:53.839 out what would happen if I did folk electronic with 0:04:53.920 --> 0:04:57.840 a really kind of action oriented music. But I didn't 0:04:57.880 --> 0:04:59.600 get a chance to get far enough into it to 0:04:59.640 --> 0:05:04.120 find out if, in fact, that is possible. Well, we 0:05:04.240 --> 0:05:07.040 have that little example there. And I don't mean to 0:05:07.080 --> 0:05:09.680 say little as in like I'm I'm dismissing it. I 0:05:09.720 --> 0:05:14.719 mean it's it's a it's a relatively um modest approach 0:05:14.800 --> 0:05:17.480 to it, and there are some really intelligent people behind it. 0:05:17.480 --> 0:05:21.520 It's a it's a group of folks from originally Yeah, 0:05:21.640 --> 0:05:26.680 so smart folks. Meanwhile, we also covered another related topic 0:05:26.800 --> 0:05:28.920 in August two thousand and fourteen, we did a show 0:05:28.960 --> 0:05:31.280 called the Future of Art, and we talked about the 0:05:31.320 --> 0:05:34.240 merger of technology and art. And today we're gonna talk 0:05:34.320 --> 0:05:38.080 about a project that's really trying to push this entire 0:05:38.160 --> 0:05:42.040 idea forward, Google Magenta, And of course it's by Google. 0:05:42.120 --> 0:05:44.880 When you're wearing your Google Fiber is coming, sir, I 0:05:44.920 --> 0:05:47.680 am that was not I didn't think about it when 0:05:47.720 --> 0:05:49.560 I put it on this morning. But yeah, I'm wearing 0:05:49.839 --> 0:05:53.760 my my Google Fiber Georgia shirt. Uh and and I 0:05:53.880 --> 0:05:57.400 understand they're laying Google Fiber along North Avenue right now 0:05:57.839 --> 0:06:00.760 as we speak. Yeah, which is which is right outside 0:06:00.760 --> 0:06:04.600 our office building. In fact, nor that all what all 0:06:04.600 --> 0:06:07.560 those police cars were escorting this morning, the Google Fiber 0:06:08.279 --> 0:06:10.760 escorting the fiber down pondstantly in Avalue right like it 0:06:10.800 --> 0:06:14.240 gets like a presidential motorcade level of welcome. One can 0:06:14.279 --> 0:06:17.000 only hope, right, I'm still I'm really holding out hope 0:06:17.000 --> 0:06:19.760 that I get Google Fiber before too long. At any rate, 0:06:20.360 --> 0:06:23.440 you might be wondering what exactly is Google Magenta. Well, 0:06:23.440 --> 0:06:27.240 it's a project that falls under the Google Brain team. 0:06:27.320 --> 0:06:30.800 That's the department within Google dedicated to using machine intelligence 0:06:30.800 --> 0:06:35.679 focused on deep learning, and in deep learning, engineers program 0:06:35.680 --> 0:06:40.000 networks of like virtual neurons that can look at data. 0:06:40.400 --> 0:06:42.720 So so let's use like images for example. Okay, the 0:06:43.040 --> 0:06:46.280 neurons can look at a picture and assess some factor 0:06:46.360 --> 0:06:49.320 of it, the shapes or the colors maybe, and then 0:06:49.600 --> 0:06:52.520 the neurons can decide what those shapes or colors were 0:06:52.520 --> 0:06:57.400 minded of and assign that decision of probability. Um, probably 0:06:57.440 --> 0:07:00.800 this is a boat. Maybe it's a banana. Most likely 0:07:00.839 --> 0:07:03.440 it's not a centipede, but it's slightly sent a petish 0:07:03.600 --> 0:07:09.200 udmule recognized boats that don't have sales. I mean, you know, 0:07:09.320 --> 0:07:12.400 it all depends on what you fed it to tell it. 0:07:12.400 --> 0:07:14.520 It was about, remember this is This is very similar 0:07:14.520 --> 0:07:16.680 to that idea we talked about, where feeding all those 0:07:16.720 --> 0:07:20.280 pictures of cats to machine learning, so that eventually the computer, 0:07:20.640 --> 0:07:23.320 without being told this is a cat, starts to learn 0:07:23.440 --> 0:07:26.480 what a cat is. It knows that it's got certain 0:07:26.520 --> 0:07:29.040 features it's really aloof, and it really doesn't care if 0:07:29.040 --> 0:07:32.640 you live or die exactly. So, so the neurons, the 0:07:32.680 --> 0:07:36.440 turons passed this information around to other neurons in their layer, 0:07:36.880 --> 0:07:39.800 and then that layer kind of compiles and passes its 0:07:39.800 --> 0:07:43.200 information up through other layers, and and the system makes 0:07:43.240 --> 0:07:46.560 increasingly educated guesses about what exactly is going on in 0:07:46.560 --> 0:07:48.520 this picture. And as you feed the system more and 0:07:48.520 --> 0:07:51.480 more pictures, it makes more and more associations with particular 0:07:51.480 --> 0:07:55.880 shapes and colors, so it learns right. And we also 0:07:55.920 --> 0:07:59.920 talked about a related uh products that came out of 0:08:00.000 --> 0:08:03.400 this group Deep dream. Oh, this is the project that 0:08:03.520 --> 0:08:06.600 turned my dog into a big old mess of caterpillars. 0:08:07.120 --> 0:08:10.239 It turned me into a big old mess of dogs. Yeah. 0:08:10.240 --> 0:08:12.120 This is This also came out of the Brain Team group. 0:08:12.160 --> 0:08:15.320 That's the Artificial Neural Network project that teaches computers how 0:08:15.360 --> 0:08:18.600 to recognize patterns out of visual data, even when the 0:08:18.600 --> 0:08:21.840 patterns aren't necessarily there. It's not it's not too different 0:08:21.880 --> 0:08:24.920 from when you look up at the clouds and you say, oh, 0:08:24.960 --> 0:08:27.680 there's a it's very like a whale. Right, So we 0:08:27.880 --> 0:08:31.480 were teaching computer programs how to hallucinate. Yeah, pretty much. 0:08:31.720 --> 0:08:34.480 And we talked about that in an episode called deep 0:08:34.640 --> 0:08:37.960 Dreaming with Google and that published in July two thousand fifteen. 0:08:38.280 --> 0:08:41.400 And and basically what is going on with deep dream 0:08:41.480 --> 0:08:43.520 here is that instead of just making a guess that 0:08:43.559 --> 0:08:47.320 a picture contained a dog, for example, um, it changed 0:08:47.360 --> 0:08:50.200 the image so that everything that appeared to be a 0:08:50.280 --> 0:08:53.280 dog in it would look more dog like. So like 0:08:53.320 --> 0:08:57.240 a layer would say, well, this this is probably a 0:08:57.320 --> 0:08:59.800 dog based on the shapes in it, So let's enhance 0:09:00.160 --> 0:09:04.200 the doggie shapes to really emphasize the dogginess. And and 0:09:04.240 --> 0:09:08.520 then as that extrapolated image is passed through Deep Dreams layers, 0:09:08.559 --> 0:09:11.520 each one emphasizes whatever it guesses is going on in 0:09:11.559 --> 0:09:14.439 the photo, and yes, frequently what it guesses is going 0:09:14.440 --> 0:09:16.880 on in the photo is a nightmareescape of dog faces 0:09:16.920 --> 0:09:18.920 as far as the eye can see. Right. So well, 0:09:18.960 --> 0:09:21.680 I think the most popular ones were definitely the animal 0:09:21.720 --> 0:09:24.800 recognition ones, but you could tweak the algorithm to recognize 0:09:24.840 --> 0:09:27.520 all kinds of stuff. And it was because that they 0:09:27.520 --> 0:09:31.000 had started with just feeding so many different animal images. 0:09:31.040 --> 0:09:33.120 That was like that was sort of their their starting 0:09:33.120 --> 0:09:35.520 point over at Google when they were training this how 0:09:35.559 --> 0:09:38.400 to recognize different visual patterns. So if a fold in 0:09:38.440 --> 0:09:43.200 your clothing looked even remotely similar to a dog, guess 0:09:43.280 --> 0:09:48.880 what you're wearing? Dogs? Now, according to arms, your arms 0:09:48.880 --> 0:09:52.120 have bugs in them, your your shirt, and your shoulders 0:09:52.200 --> 0:09:55.560 or dogheads, your dog actually is not a dog but 0:09:55.600 --> 0:09:58.200 a bunch of cats taped together. Yeah. I like that. 0:09:58.520 --> 0:10:02.200 Everything became very dog lee for a while. But one 0:10:02.200 --> 0:10:04.280 of the other projects under the Brain team is one 0:10:04.320 --> 0:10:06.800 we're going to mention a little bit later called TensorFlow, 0:10:06.840 --> 0:10:09.319 which is a machine learning engine, and it's an open 0:10:09.360 --> 0:10:13.440 source project, meaning lots of people anyone really can access 0:10:13.440 --> 0:10:16.720 those tools and not just access them, but tweak them, 0:10:16.760 --> 0:10:20.480 improve them, evolve them, and and grow them, and that's 0:10:20.480 --> 0:10:25.120 really an interesting and potentially exciting development in machine learning. Yeah, 0:10:25.160 --> 0:10:28.360 but you can see how pro projects like this might 0:10:28.480 --> 0:10:33.199 be sort of evolving toward ultimately generative powers in AI, 0:10:33.320 --> 0:10:37.560 not just recognition and modification of visual images and other 0:10:37.600 --> 0:10:42.319 types of sensory input and data, but but actually building 0:10:42.400 --> 0:10:45.559 things from the ground up, recognizing patterns and saying I've 0:10:45.559 --> 0:10:47.600 got enough sense of what the patterns are that i 0:10:47.600 --> 0:10:50.240 can make one on my own, right, exactly like you, 0:10:50.240 --> 0:10:53.520 You no longer are taking a a uh, you know, 0:10:53.960 --> 0:10:58.199 an image or a concept and then saying, here, enhance 0:10:58.320 --> 0:11:01.680 this or a this in some way or recognize it. 0:11:01.679 --> 0:11:03.560 And you're now saying, hey, you know what one of 0:11:03.559 --> 0:11:06.520 those happens to look like? Make one? Yeah, And that's 0:11:06.920 --> 0:11:09.559 that's a big leap, right, that's a huge leap. And 0:11:09.920 --> 0:11:12.240 you can do the same thing you somebody could say, 0:11:12.280 --> 0:11:14.200 tell me the plot of a James Bond movie that 0:11:14.240 --> 0:11:16.600 doesn't exist. You've seen enough of them. You know, there's 0:11:16.600 --> 0:11:19.520 a standard pattern. You gotta have the gadgets, you've got 0:11:19.640 --> 0:11:24.160 to have the you know the yeah, yeah, yeah, you 0:11:24.240 --> 0:11:26.520 gotta have a pit of sharks or something like. You 0:11:26.520 --> 0:11:28.360 you can put the pieces together. You gotta have at 0:11:28.440 --> 0:11:31.960 least one one lady who is pretty much a good 0:11:31.960 --> 0:11:33.679 guy and one lady who's pretty much a bad guy, 0:11:33.679 --> 0:11:35.680 and Bond's gotta mess with both of them. You know, 0:11:35.720 --> 0:11:40.679 there's there's certain rules that you probably both what's that 0:11:41.040 --> 0:11:45.079 messed with? Hey, I'm being very uh, you know, being 0:11:45.160 --> 0:11:49.400 very family friendly here, so like James Bond, Yes, exactly, 0:11:49.480 --> 0:11:53.200 James Bond, so family friendly. So the Magenta project is 0:11:53.240 --> 0:11:58.439 aimed at developing artificial intelligence capable of actually creating art, 0:11:58.679 --> 0:12:02.719 not not just altering something so that it looks like art, 0:12:02.920 --> 0:12:06.360 but to create art, both visual art and music. Now, 0:12:06.400 --> 0:12:09.079 the official launch date of Magenta is June one, two 0:12:09.080 --> 0:12:12.320 thousand sixteen. We are recording this on May twenty six, 0:12:12.480 --> 0:12:15.000 two thousand sixteen, so it has not launched as of 0:12:15.040 --> 0:12:17.800 the time we're recording this, UM, but we wanted to 0:12:17.840 --> 0:12:20.400 kind of talk about it, and it was just recently 0:12:20.480 --> 0:12:24.439 announced to the world, although not officially unveiled. Douglas k 0:12:24.480 --> 0:12:28.120 who's working on the project, announced Magenta at Mogue Fest 0:12:29.040 --> 0:12:33.959 I assume named after the Mogue synthesizer which everybody lives 0:12:34.360 --> 0:12:36.920 and um that's actually a music and technology festival that 0:12:36.920 --> 0:12:39.000 takes place in North Carolina, so it was not too 0:12:39.040 --> 0:12:42.760 far away, and also stressed that while they're working on 0:12:42.800 --> 0:12:45.120 this project and while they have high hopes for it, 0:12:45.320 --> 0:12:48.520 he says, AI is still a very long way from 0:12:48.520 --> 0:12:52.360 creating long narrative arcs. So it's not like this is 0:12:52.400 --> 0:12:54.720 going to be, you know, within a year, we're going 0:12:54.760 --> 0:12:57.280 to have computers righting the next great American novel or 0:12:57.320 --> 0:12:59.560 anything along those lines, but that this is the first 0:12:59.559 --> 0:13:04.120 step to word computers making a creative problem solving and 0:13:04.280 --> 0:13:06.839 moving to a point where that problem solving isn't about 0:13:07.000 --> 0:13:11.480 tackling a question, but about creating something new, like music 0:13:11.640 --> 0:13:15.240 or a painting or anything along those lines, or even video. Well, 0:13:15.280 --> 0:13:18.760 I mean, I think something literary and long and coherent 0:13:18.920 --> 0:13:21.480 like a novel would would be one of the most 0:13:21.480 --> 0:13:25.680 difficult things because that involves just the most uh what 0:13:25.800 --> 0:13:29.720 might you call it, semantically diverse array of things that 0:13:29.760 --> 0:13:32.400 you're working with. I mean, in a lot of moving parts, 0:13:33.320 --> 0:13:35.439 not just not just the words that have to make 0:13:35.480 --> 0:13:38.719 sense and sentences and in paragraphs and in chapters, but 0:13:38.800 --> 0:13:42.480 also character development, character motivation. I mean, it's not even 0:13:42.880 --> 0:13:47.040 that doesn't even make sense to a computer, world building, etcetera. Right, Yeah, exactly. 0:13:47.040 --> 0:13:48.840 There are a lot of things, a lot of elements 0:13:48.880 --> 0:13:53.360 in creating a long narrative that are would take a 0:13:53.400 --> 0:13:56.280 long long time to teach a computer. What does this mean? 0:13:56.440 --> 0:13:58.280 I mean you can imagine that a book written by 0:13:58.320 --> 0:14:00.920 a computer before it has a full grasp on that, 0:14:01.160 --> 0:14:03.840 it could end up being incredibly dull. You're just you're 0:14:03.840 --> 0:14:07.720 reading a very mundane account of a person or or 0:14:07.720 --> 0:14:10.560 it could be the opposite. It could be like well, 0:14:10.600 --> 0:14:12.520 it could be like a Dan Brown novel where every 0:14:12.559 --> 0:14:15.240 page ends with a cliffhanger and you think I need 0:14:15.320 --> 0:14:19.720 to break you know. I was looking up Douglas X 0:14:19.720 --> 0:14:21.560 so I was reading a little about him on his 0:14:21.840 --> 0:14:24.880 Google research page, and one of the things he had 0:14:24.880 --> 0:14:30.200 previously worked on was was content. What would you call it? 0:14:30.360 --> 0:14:33.800 Uh oh, I'm losing the word for it. He worked 0:14:33.800 --> 0:14:37.440 on Google Play music delivering you the kind of music 0:14:37.560 --> 0:14:41.280 you would want. There's a term for that, curation. Curation sure, yeah, 0:14:41.320 --> 0:14:44.440 based on yeah, yeah, yeah, based on what you like 0:14:44.560 --> 0:14:47.320 to listen to. Figuring out, Okay, what's the other type 0:14:47.360 --> 0:14:50.200 of music you haven't heard yet that fits in with 0:14:50.280 --> 0:14:53.360 the profile of the stuff that you like, which is, 0:14:53.680 --> 0:14:56.000 you know, that's kind of a tough job, because how 0:14:56.000 --> 0:14:58.600 does the computer know that one song? Oh you know, 0:14:58.880 --> 0:15:02.400 if Jonathan really likes this song by they might be giants. 0:15:02.480 --> 0:15:05.920 He'll probably really also like this song by Slayer. Yeah 0:15:06.080 --> 0:15:08.200 I think that. Yeah, they're like, hey, like like you 0:15:08.240 --> 0:15:11.200 like rock and roll Joe, so probably you like this 0:15:11.400 --> 0:15:15.920 Nickelback song which is rock and roll. Yes, yeah, yes, 0:15:16.080 --> 0:15:18.720 So at least with the Music Genome Project, which is 0:15:18.760 --> 0:15:21.600 what Pandora is based off of, the way that works 0:15:21.720 --> 0:15:25.360 is they have human beings who meta tag every song 0:15:25.400 --> 0:15:29.080 with every kind of descriptor that would be relevant to 0:15:29.200 --> 0:15:33.400 that particular song, and then the algorithm starts looking for 0:15:33.480 --> 0:15:37.440 other songs that have several of those same meta tags 0:15:37.480 --> 0:15:39.840 associated with it and say yeah, yeah, and so it's 0:15:39.880 --> 0:15:43.080 going like you like melancholy guitar solos because you listened 0:15:43.160 --> 0:15:45.800 to the Decembrists, so probably you're going to like the 0:15:45.880 --> 0:15:49.720 dashboard confessionals, right, So that that may be similar to 0:15:49.760 --> 0:15:52.400 the way Google Play Music does it. I don't know, 0:15:52.440 --> 0:15:56.400 because I don't know. He made it sound more like 0:15:56.480 --> 0:15:59.200 that this was a more automated process. That's really interesting, 0:15:59.280 --> 0:16:02.520 more difficult. Yeah, yeah, you're not having a human hold 0:16:02.560 --> 0:16:05.200 your hand taking you through Okay, now this is what 0:16:05.320 --> 0:16:11.240 moody sounds like. This is melancholy, right, that's really interesting. Well. 0:16:12.000 --> 0:16:15.760 At at the mog Fest conference, Magenta team member Adam 0:16:15.880 --> 0:16:19.280 Roberts showed off a digital synthesizer program where he could 0:16:19.320 --> 0:16:21.240 feed a few musical notes into it. I think it 0:16:21.240 --> 0:16:24.520 was a sequence of four notes, and then allowed the 0:16:24.560 --> 0:16:27.800 program to build a melody off those basic notes. And 0:16:27.840 --> 0:16:30.600 what I'm picturing this kind of going off the deep 0:16:30.680 --> 0:16:33.160 dream concept, is you know like, Okay, well these notes 0:16:33.680 --> 0:16:36.240 sound like the beginning of a big band slow dance, 0:16:36.240 --> 0:16:38.080 So I'll just add more notes to make it look 0:16:38.280 --> 0:16:40.760 more like that. Sound more like that. Yeah. The the 0:16:40.920 --> 0:16:44.560 example that they showed in the in the actual festival, 0:16:44.560 --> 0:16:47.760 when I when I watched it, I thought, ah, they 0:16:47.800 --> 0:16:51.360 picked up bad for initial notes. It just because you 0:16:51.440 --> 0:16:53.360 listen to it and you're like, to me, that sounds 0:16:53.400 --> 0:16:57.600 like just someone aimlessly plunking on a synthesizer keyboard. It 0:16:57.600 --> 0:17:01.040 didn't sound like someone actually creating melody. But I would 0:17:01.120 --> 0:17:05.119 argue that you're really in this case, the way this works, 0:17:05.680 --> 0:17:08.600 the the final tune is really only going to be 0:17:08.640 --> 0:17:13.000 as good as the initial input you give to the computer. 0:17:13.160 --> 0:17:16.840 Because it's not creating it out of whole cloth. It's 0:17:16.880 --> 0:17:19.760 taking a foundation and then building upon it. If the 0:17:19.800 --> 0:17:22.240 foundation is faulty, then you can't really expect the rest 0:17:22.280 --> 0:17:26.040 of it to be awesome. Um So, but maybe that 0:17:26.119 --> 0:17:28.440 was just me. I also don't have the best ear 0:17:28.760 --> 0:17:33.080 so perhaps I'm being particularly harsh. Also, this this is 0:17:33.119 --> 0:17:35.760 an early prototype that he's showing off, right, Yeah, and 0:17:35.800 --> 0:17:38.120 maybe that the what we see on June one will 0:17:38.160 --> 0:17:43.400 be a more comprehensive demonstration of the abilities of Magenta. 0:17:43.480 --> 0:17:49.440 And keep in mind the Magenta we had it composed 0:17:49.440 --> 0:17:54.119 this song and destroy these four countries? What what did 0:17:54.240 --> 0:17:57.719 you do? Now? Um? We think? Uh. For one thing, 0:17:57.760 --> 0:18:00.359 Magenta is supposed to be an ongoing project, right. This 0:18:00.440 --> 0:18:03.520 is not something that's a fully fleshed out product and 0:18:03.600 --> 0:18:06.080 they're going to reveal it on June one for people 0:18:06.160 --> 0:18:10.399 to play with. It's more like, here's the concept behind 0:18:10.400 --> 0:18:12.960 the project, here's how we're going to try and accomplish 0:18:13.000 --> 0:18:16.159 our goals. Here's where we are now. That's more likely 0:18:16.280 --> 0:18:20.160 to be the announcement on June one. So uh. Ex 0:18:20.280 --> 0:18:24.520 hope is that by feeding enough musical information to Magenta, 0:18:24.720 --> 0:18:28.000 enough songs, in other words, it will be able to 0:18:28.000 --> 0:18:31.359 produce its own music that is esthetically pleasing to human 0:18:31.440 --> 0:18:36.639 like people persons, and that those yeah, I mean that 0:18:36.760 --> 0:18:39.520 my list is pretty small, but I know a couple. 0:18:39.880 --> 0:18:42.800 So the program first has to learn what makes music work? 0:18:42.880 --> 0:18:46.720 What are the rules of music? So what are the 0:18:46.720 --> 0:18:49.280 sort of things that we like to listen to? And 0:18:49.320 --> 0:18:51.879 once you know those rules, when is it okay or 0:18:51.960 --> 0:18:55.639 even preferable to break those rules? When are when is 0:18:55.640 --> 0:18:59.240 it all right to stray from the conventions of any 0:18:59.240 --> 0:19:02.680 particular musical genre and do so in a way that's 0:19:02.760 --> 0:19:06.560 interesting and and maybe it's pleasing, maybe it's maybe it's 0:19:06.600 --> 0:19:09.200 not pleasing, but it's the sort of thing that catches 0:19:09.240 --> 0:19:12.639 your attention and that's what makes the music really stand 0:19:12.640 --> 0:19:16.280 out to you. Right, So these are not easy concepts, 0:19:16.320 --> 0:19:19.080 even for musicians, like human musicians it you know, it 0:19:19.080 --> 0:19:22.000 can take years of study to really understand musical theory 0:19:22.480 --> 0:19:27.280 and be able to craft something that is most likely 0:19:27.320 --> 0:19:30.600 to evoke the reaction your hope to get from your audience. 0:19:31.160 --> 0:19:33.920 You know, I mean, I know, I've written a couple 0:19:33.960 --> 0:19:38.480 of songs. They're terrible, uh, rather than any rate. So 0:19:38.640 --> 0:19:40.800 music is just the first type of art the Magenta 0:19:40.880 --> 0:19:43.159 is going to tackle. They're going to actually use the 0:19:43.200 --> 0:19:45.560 same sort of approach to generate images and perhaps even 0:19:45.720 --> 0:19:49.119 video in the future that they mentioned eventually text too. Yeah, 0:19:49.160 --> 0:19:51.639 so again getting to that point where maybe we can 0:19:51.640 --> 0:19:53.479 get to that narrative arc. And this also kind of 0:19:53.480 --> 0:19:55.880 feeds into when we were talking of our I don't 0:19:55.880 --> 0:19:57.760 know if we've ever talked about on the podcast, but 0:19:57.840 --> 0:20:01.320 Google recently in the news is uh, a lot of 0:20:01.320 --> 0:20:03.920 people poked fun at Google because it was feeding romance 0:20:03.960 --> 0:20:07.640 novels too. It's um digital assistant, so it would learn 0:20:07.680 --> 0:20:11.560 better about how people converse. The important thing being that 0:20:11.880 --> 0:20:16.159 romance novels tend to follow a very similar uh uh 0:20:16.720 --> 0:20:20.480 pattern right, you're very they're very formulaic, but different romance 0:20:20.520 --> 0:20:23.399 novels say the same thing in different ways. And so 0:20:23.960 --> 0:20:26.600 the hope is that by feeding this kind of information, 0:20:26.640 --> 0:20:28.880 and romance novels were just one genre that we're fed 0:20:28.880 --> 0:20:30.680 to it, but everyone focused on it because of course 0:20:30.680 --> 0:20:33.840 it's funny. You know, this idea that your digital assistant 0:20:34.240 --> 0:20:38.480 is going to be making some very sassy recommendations to 0:20:38.520 --> 0:20:41.600 you or explain the weather in ways that are probably inappropriate. 0:20:41.920 --> 0:20:44.240 But the idea being that by feeding all this information 0:20:44.320 --> 0:20:47.280 using machine learning, that you would be able to get 0:20:47.400 --> 0:20:51.600 your finished product to be more capable of interacting with 0:20:51.640 --> 0:20:55.480 people using natural language, very similar to what Magenta is doing, 0:20:55.520 --> 0:20:57.800 except in that case it's music and art. Right. You're 0:20:57.840 --> 0:21:01.040 feeding more and more music in to Magenta so that 0:21:01.119 --> 0:21:04.879 it has a quote unquote understanding of what music is 0:21:04.920 --> 0:21:08.520 and can more likely produce something that is similar to 0:21:08.560 --> 0:21:11.359 what a human would make without it actually just copying 0:21:12.080 --> 0:21:15.280 something that a human has already made. Yeah, I mean 0:21:15.320 --> 0:21:19.720 I wonder sort of what the what the end goal, like, 0:21:19.760 --> 0:21:22.720 what the expectation is here, because like, on one hand, 0:21:22.760 --> 0:21:25.359 if we're to believe what's supposedly going on in the 0:21:25.359 --> 0:21:28.240 back end that creates these tracks we listen to a 0:21:28.359 --> 0:21:31.359 juke deck, I feel like, here's the system that is 0:21:31.400 --> 0:21:36.160 already creating perfectly passable music. Again, like I said, it's 0:21:36.160 --> 0:21:39.520 it's not amazing. So I'm wondering, is is Magenta aiming 0:21:39.560 --> 0:21:44.080 to create music that's really going to be amazing and 0:21:44.119 --> 0:21:47.399 people will be like, wow, I love that song. I 0:21:47.440 --> 0:21:50.879 think Magenta's not to put words into the mouths of 0:21:50.920 --> 0:21:54.760 the people in the project, but I think Magenta, by 0:21:55.000 --> 0:22:01.240 using a very specific goal, is really out driving machine 0:22:01.320 --> 0:22:04.000 learning further. Yeah, so I agree. I think that the 0:22:04.119 --> 0:22:07.199 art is absolutely secondary and kind of the headline grabber, 0:22:07.440 --> 0:22:11.000 right um, and and that right the other applications like 0:22:11.080 --> 0:22:13.320 in a voice recognition or something like that, or what 0:22:13.359 --> 0:22:15.840 they're really aiming for, right So. So for a great 0:22:15.840 --> 0:22:18.080 example of this would be looking at the private space 0:22:18.160 --> 0:22:22.240 industry and saying, you know, you've got some pretty big, 0:22:22.280 --> 0:22:25.360 big ideas, how do you focus that in a way 0:22:25.359 --> 0:22:28.560