Chocolate Chicken Chicken Cake

Kind of haunting, abstract, yes, but beautiful too. And crucially, when I read this, I felt, as you just did. I hope that it is beautiful. I found it evocative of experience that I've had it in the past. I found it nostalgic um and I'm like, oh my god, I'm having real human being emotions. That was filmmaker Oscar Sharp, and that poem wasn't written by him or by anyone else. It was written by a computer, a machine poet. We're more and more worried about robots coming to take our jobs.

And though perhaps few would regret less hips to poets shambly around Brooklyn, somehow machines in the creative world are especially uncanny, even frightening, because poetry and music and humor are supposed to be the things that define our humanity, aren't they. In this episode, we look at how AI is being used in the creative arts, and in doing so, we understand a lot more about how this often intimidating technology actually works. I'm Moza Lush and welcome to Sleepwalkers. Hi,

Hi Karen So did Oscar's poems spring your delight? I would have preferred it in your British act? Then I think, well, I can't blame you for that. It did remind me of Um. My friend is in a band called the x Ambassadors and they partnered with this producer called Alex the Kid who was asked by IBM to make a song using Watson. And the song is not bad. The

song is not bad. Um, the sounds good. But basically, the way in which they used Watson was that they crunched the twenty six thousand songs from the top one charts. It's from over what time period, like over a few years presume, I don't know, But the point is that they used them to discover patterns in the songs. What makes the top hundred song. Basically that's right, and then that's reproduce it right, which is interesting because I think anybody could kind of tell you what makes the top

on hundred song right. They're called earworms. But when you think about a data set of twenty six thousand, like, no human being can listen to that many songs and do any productive after they hear it. So in this episode, we're going to look at AI and art and different kinds of fields to really understand how computers crunched data

to crack open this creative code. But first I want to go back to the algorithm who wrote the poem we heard at the beginning of the episode, because that algorithm also wrote the film that Steven Spielberg of the machine learning world, and the algorithm is called Benjamin. So we're going to meet Benjamin and a few people not named Benjamin. I was a speech trator for John Kerry, Tim Gaitner, and Barack Obama, um, not in that order, And I'm essentially a ghost writer and a photographer who

learned to code one giant Frankenstein monster. That's Ross Goodwin and that Frankenstein monster it goes by Benjamin and it's the work of Ross Goodwin and Oscar Sharp, who you'll remember from that poetry reading, but neither of them actually named Benjamin. Well, they named itself, or rather, there was a piece of paper that came out of it that said my name is Benjamin on it. I read that in response to a question that was put to it, and a room full of people went, oh, a program

that names itself is rather uncanny. And Oscar had been chasing that uncanny nous ever since he was at n y U for graduate school. Whenever I met anyone who good program, I would grab them by the pels and yell into the face. Can you can you build something that can write like people talk in some way? And one day in class Oscar notices there's this sneakerhead and he's sitting on his laptop and his laptop is writing

without him touching it. And I'm like, oh, so we go, So we go for coffee, and that gut coffee was. It was a lengthy coffee. We're still having that coffee. We're still having such a cold coffee. By now, you might not believe it if it happened in a film, but Oscar had stumbled on exactly the person he was looking for. Oscar came to me and he said, I want to make a movie from a computer generated screenplay. And I said, you know, of course that sounds amazing.

Let's do it. But let's figure out how we're going to generate the screenplay, because that's a nuanced process with lots of stabs, and we need to consider like every part it. So Oscar volunteered himself to teach me all the things about storytelling and narrative and filmmaking. He turned me onto like Vladimir prop Joseph campbell Um, all these

theories of storytelling, and so they begin to experiment. I tried a bunch of prototypes that used like various structures that had been postulated by these theorists over time, and the output was not interesting. Despite following the rules laid out by narrative theorists, Ross couldn't get anything good just

telling his programs what a story should contain. So a year passes, Oscar moves to l A and when I get this email from Ross and it's the results in one of those experiments that he wants me to read, and read it he did. Rossity mailed the poem from the beginning of the episode, the room is blown away from the door and the stones are beginning to shine. I immediately was like, oh my god, I don't know

how he's doing this. But he said, I don't know what technology you're using right now, but can we it for screenplay? And so they did, and not just a screenplay, they actually produced a short film called Sunspring and they even got Thomas middle Ditch, the lead on Hbos Silaken Valley, to star in it. Principle is completely constructed. Of the same time, it's all about you. To be true, you didn't even watch the movie with the rest of the base.

I don't know, I don't care. I know it's a consequence whatever you need to know about the presence of the story. I'm a little bit of a boy on the floor. So what do you think, Carol? It kind of reminds me of when my parents used to take me to like a bad production of Macbeth or as you like it. Traumatic. You're there and you're seven or eight, and you want to understand what's going on, and so you kind of pay as close attention as you possibly can to what the actors are doing because you have

no idea what the dialogue means. Yeah, I mean that to me is quite impressive because a machine can create something which has enough of the elements in common the film that we can talk about a real film. You can't say it's not a film absolutely. Of course. What's different is it didn't take Benjamin very long at all to make it. Once you press the button fraction of a second there was a couple of seconds perpase, maybe maybe a couple of seconds total, actually a fraction of

a second per page. That's right. After months of agonizing over centuries of storytelling theory, the final output only took a couple of seconds. So what was Ross's breakthrough? To understand we turned to one of the most famous AI scientists in the world, Sebastian Throne. Recently, something magical had happened recently. The feat has discovered was called machine learning. With AI, computers can now find their own rules. They

are called neural networks. They're comprised of hundreds of millions of little vase sample processing units, and those units are modeled after what a neurons do in our physical brains. You just give them examples, very much like the way we we waste children. We don't give our children rules for every contingency in life. In the first eight years of education. We let them learn, They experience the world,

and they loan behold. They make their own rules. And we are now in the world where computers can do the same thing. And this means machine learning can be used in all kinds of different fields. Sebastian himself applied the technology at Google, where he led the initial development of their self driving car. When you want to read a book, a book on like what the car should do in every situation, that rule book is really complicated and it can promise you no matter how many years

you spent writing it, it's not gonna work. But when you give the machine the ability to learn its own rules, it is actually able to surpass how people can drive. We'll hear more from Sebastian later, but machine learning mL is the engine that drives almost all of the excitement about AI today, from identifying targets on the battlefield to understanding genetic diseases. And it's also what allowed Ross and

Oscar to create a usable movie script. Rather than laying down storytelling rules, they simply showed Benjamin hundreds of examples and the algorithm found patterns and learned for itself more sleepwalkers. After the break, you're like, oh, did we essentially we teach this algorithm anything else about screenplay other than just putting in a bunch of screenplays, right, And that's the

way that machine learning works. What is happening in a deep learning algorithm of this kind is it's building an extraordinarily complicated mathematical formula by reading all of this stuff over and over again, like the auto complete on your phone. The neural that is actually sampling from a probability distribution of which letters, bass or punctuation become next. So the script for sun Spring was essentially the most mathematically probable Sci Fi script except Ross and Oscar did have one

important lever of creative control. The other of parameters that you're probably wondering about, there's one called the temperature is the riskiness of those next letter predictions. What Ross is describing is almost like a dial for creativity. Turn it up to a really high temperature, and the neural net is going to be extra creative and start making up words, babbling at a very high temperature. It's essentially drunk. Low temperature, it's going to be very repetitive and possible even begin

to plagiarize its source material. So it'll be very repetitive. It'll be like the streets and the streets, and the streets and the streets. It's essentially went working for network television. Yeah, exactly. So we wanted it to be sort of in the middle. In the middle is where we found the best output and the most I think usable output, and Sunspring was born. So Benjamin Ross and Oscar right together now they write poetry and movies and sometimes what Benjamin spits out is good.

Often they have to sift through it to find the best stuff. But he's prolific and he never ever suffers from writers. Look, so Kara was telling us earlier about Alex the kidd and using AI to make music, and that's something I want to understand a bit more about. So Julian went on a little bit of an expedition. Yes, I did. I've been seeing a lot of articles lately about AI and the arts, and I've been pretty curious

about music specifically. We might take it for granted, but music is this primal emotional thing that's been with us forever. It might even predate language. But now Warner Music Group made history in April two nineteen, is the first major label to sign an AI to a record deal. Yeah, they signed this bot called Endel, which makes ambient noises based on where you are and what the weather is

and what time of day it is. When I think of this kind of music, I think of those Spotify playlists like Peaceful Piano and Blissed Out Dinner Party, which would become extremely popular. It's not the same thing as Beyonce, No, definitely not. But Warner Music Group signed Endl to generate twenty albums of ambient music. And now that we live in a world where aies can get record deals, what does this mean for artists? What does this mean for

even just music as we know it? Well, in my quest to find out, I visited this company called Amper. My name is Drew Silverstein. I am the co founder and CEO of Amper Music. Amper is an AI music company that Drew says will enable anyone to create music. In fact, the only things you need to know are the genre of music you want to create, the mood you'd like to convey, and the length of your piece

of music. That's all you know. You can create a brand new, unique piece of music in a matter of seconds. So the big question is should musicians worry about computers taking their job? Well, let's try it and see. So what do you want to do? Cinematic, documentary, folk cinematic cinematic, minimal percussion or quirky percussion. It's rendering a song right now. And here we go. We've got something. I'm were deep in the forest of Nicaragua. There's a breed of jaguar.

You might have heard of it. It's called the take a Killer panther. Look, here's the thing. I don't know what the difference between that and music is. I really don't. Yeah, so you're woud Yeah, all right, So there's that. And this isn't the only AI music app out there. Another major player is called Magenta, and big surprise, they're at Google. Magenta are using AI to create a ton of new tools.

From a piano genie that makes it impossible to play bad notes to something that can generate drum loops, or something that can even play piano duets with you. You can even translate raw audio to a piano score. Raw audio, like if I play just something raw on the piano, raw audio like oh, literal raw audio, literal raw audio. And Magenta has also trained a neural network just like Ross and Oscar, only instead of sci fi scripts, they

trained on over four hundred performances by skilled pianists. They fed it into the neural network and let me play one of the piano experts. First. This is a real piano player, all right, so nice? Right? Yeah, okay, ready for the AI. What that's a computer that was all a computer. I didn't ever play a human one. That's a computer that was trained by a human playing piano. And then how do you make a computer come up

with that? Right? So even though it's not a screenplay, it's still data that you can feed a neural network with to find patterns. And in this case, Magenta used a data set from the Yamahai Piano competition. So human pianists played on these digital keyboards which recorded the nuances of their performance, like how long they hit notes, and it recorded all that information into a digital score that a computer could interpret. And we've actually had that technology

for a while now, it's called MIDI. But training and on network on the data is new. See. The thing that I come back to is that a computer doesn't know it's playing music, so much of watching a musical performance is knowing that this is coming from someone who is emoting. Right, Yeah, there's actually there's an emotional communication happening, right, that's right. I do think though the future is not

rejecting this. It's better to imagine what would Stravinsky have done with this kind of technology, because Stravinsky is still a musical genius. Right, Yeah, Definitely it's cool to listen to those musical examples of machine learning because you can really hear how the algorithm is reinterpreting existing material. Of course, listening to the output is one thing. Tasting it is

quite another. The problem was that somebody had told me that they had made the recipe for Stan, that it was good, and what it was as a recipe called chocolate baked and serves. That's Janelle Shane. She's a research scientist and the author of a blog called AI Weirdness. She's talking about a recipe written by Ai that she actually cooked and eight. It starts out as a perfectly ordinary, flowerless chocolate brownie all the way until the very last ingredient,

which is a cup of horse badish. I knew I was in trouble when I opened the oven door and my eyes just started watering. It was yeah, it was terrible. On her blog, Jenelle experiments with putting AI to a range of tasks, from writing new pickup lines to naming Halloween costumes, and often her experiments with machine learning are pretty revealing about us. It plays into this thought experiment,

what would an alien think of our world? It takes something that's very ordinary and mixes it up into this thing that sounds like the original, but the meaning has been completely changed. Chopped whipping cream may be an ingredient and fold water, enrolled it into cubes, or spread the butter in the refrigerator. That's another direction that came up with. Remember Ross and Oscar playing with the creativity setting for

their scripts. Janelle plays with herbot's temperature too, so I can turn it up and the neural net may choose its second best or third best guess as to what letter comes next. And if I turn the creativity all the way down, then everything maybe something like the the the or recipes may be just you know, one teaspoon of vanilla over and over and over again, because that's just a very likely ingredient. It's really interesting with the recipes to turn down the creativity and see what it

comes up with as the most quintessential recipes. At the lowest setting, you may not get hole Strandish brownies, but you do get a clear picture of what we eat and who we are. I look at what kinds of recipe titles that comes up with. There are things like chocolate chicken chicken cake, and another one that's chocolate chocolate chocolate chocolate cake. And there was a lot of cheese in these recipes too, so it's kind of revealing about what sorts of things we cook with. Then we like chocolate,

cheese and chicken apparently. But then I did the same experiment with recipes from Bone Appetite, and then the most common ingredients that kept using were cilantro and pomegranate juice. So these algorithms essentially hold up a mirror to the data sets that we give them. They do, yeah, they reflect the data sets back to us in really weird ways, and they can absolutely pick up whatever bias there is

in a input data set. And I think what we're discovering is just how prevalent that bias is and how easy it is for neural networks to latch onto that bias and copy it as a handy tool toward copying whatever where the humans are doing. They say that the way to a person's heart is through their stomach. But Janelle didn't stop at chocolate, bakes and surfs. She's also turned AI onto some more direct roots. I really liked

the pickup lines. And there are all these puns and all this wordplay that it didn't have any way to grab hold of and figure out how to use. But I think what it produced this sort of charming surrealism and kind of garble nonsensical. I think it's an improvement on every single one of the originals. My very favorite one is you look like a thing and I love you. You are so beautiful that you make me feel better to see you. Or you must be a tringle because

you're the only thing here. Are you a camera? Because I want to see the most beautiful than you. Yeah, I'll definitely lie with you. No one's have a used real pickup line on me? Use one on you know? Do I look like someone who would receive a pickup Well, here's one of them. I don't know you. That's good a lot of girls are into that. Are you a candle? Because you're so hard of the looks with you. So in effect, what the algorithm is doing is highlighting patterns

in the data. I mean, there's sound structural like pickup lines about the words themselves don't make any sense. The machines are reflecting their creators and spitting back something which resembles pick up lines and makes us think a little bit more carefully about what a pickup line is. And while training Benjamin Ross and Oscar found the same thing as the algorithm learned patterns revealed bias present in our cinema.

When you train an algorithm like Benjamin on millions and millions in this case of synopsis from the Internet of the movies, the synopsis that come out have certain patterns in them. For example, they mentioned men full times more often than they mentioned women. But you you learn other things about it than that. You learned that the most common phrase in the in the output is a young man in a small town. So what does a filmmaker like Oscar learn from this? I used to call this

project the average movie projects. And the reason I called it that is the theory was for me, if you could make the right kind of algorithm that the movie that you would make that would be the theoretically perfect movie would also be the most boring movie ever made, and that it would. It would it would be by definition all of the things that were the most clich because that's what cliche means, is the thing that that you can rely on to work. And why do that?

Because the thing I'm most interested in is doing the thing that we haven't done yet. I want to move the form forward. Seeing all of these biases and assumptions that are baked into our movies and our snacks doesn't mean we're doomed to repeat them. In fact, the awareness can be liberating. That's what's helped me, I think, is seeing Benjamin's capacity to show me more directly what it is that are our patents, are our habits, and then I can ask more easily how to move forward from that.

We'll get there after the break. So we've heard about Janelle Shane using AI to reveal bias, and Ross and Oscar using it to help them think more creatively about filmmaking as well as how it can be applied to music. And that's the great promise of AI. We may worry about replacing jobs, but it can augment our lives in so many ways. At least that's how Sebastian Throne sees it. I would say the term AI is a bit deceptive because it sets up computers to be on equal power

with people. I see it to be stronger where we are weak, and weaker where you're strong. It's not a technology that will replace us, as it's not really empowers but what might that empowerment look like beyond bias detection and piano playing well. In seventeen, Sebastian published a paper in Nature on using AI to diagnose skin cancer using

just an iPhone. So in medicine, you can think of your iPhone that can find skin cancer as turning regular physicians or anybody in the world into an expert on day one, because now they have the superpower to be able to distinguish something that previously would have taken tens of years to learn. The same is true for the self driving car. Now children can drive and and and blind people can drive, head blind people drive around and

self driving cars. So for me, the real opportunities to use the eye to extract the knowledge from some human experts that are well trained and transpose this knowledge to other brains that people not so well trained. This is what I personally find so fascinating about AI and a big reason I wanted to do Sleepwalkers. The same technology underlies rossan Oscar's films, Janelle's recipes and self driving cars

and cancer diagnostics and so much more. The training for skin cancer detection or cancer detection radeology and the training for the self dime car amazingly similar. In both cases, what you do is you compile the data set typically hundreds of thousands up to hundreds of millions of images. In skin cancer, we use biopsies. We had a database of a hundred twenty nine thousand images that a lab

had biopsied and provided. In the self driving car, you could be as easy as having a human driver provide inputs with their student veel and the and the break as to what the right thing is to do, and then the network mimics human behavior, it mimics the diagnostics of a physician, or it mimics the style of a driver. The underlying albums are amazingly similar. What makes this moment all the more interesting is that AI is in the process of being consumerized like Sebastion said, your iPhone can

diagnose skin cancer. Self driving cars are already on the roads, and as these tools become more and more accessible, society will start to change. These technologies become closer and closer to us. The fact that you carry yourself phone is a big deal. You might not see this way, but what it does it puts the computer seamlessly into your life. You're texting app, your SMS is so close to you that you can now talk to people thousands of miles of a on a button press or on a microphone.

That makes you effectively super human without the actual physical implant. But when it comes to AI, for now, leading edge algorithms are off limits to those of us who can't code or who don't have the means to learn. Oscar wouldn't have been able to make Sunspring without Rosses technical expertise, but that's all starting to change, Julian, you spoke with

somebody making AI more accessible. Yeah, while we were looking into AI and music, we came across Runway m L their lab based in Bushwick, and they feel strongly about letting more people into work with AI Creatively. I spoke with Christo bal Valence Weela, the co founder of Runway, which is basically like the Adobe Creative Suite for AI.

So think of a program that looks like Photoshop. They're adding tons of different AI models to the Runway app, where instead of having to know how to code, you can just manipulate some sliders and dials and still have

a I generate something. If we really think of this the game changing technology that will impact us for like years to come, we need to have more people from different backgrounds and disciplines jumping to that discussion and proposing ways of looking at algorithms that those researchers and those scientists are not thinking of. This is gonna impact us, and it's going to change the way we see not just the world, but ourselves. Not just ourselves, but how

we think about our creativity. And the list of things that Runway can help you do is frankly crazy. Will post some on our Instagram at Sleepwalkers podcast, but just one example, you can take video of anyone adds and have their body copy the poses that you make in your webcam. So Krystoval actually tweeted one of him controlling Stephen Colbert's body on The Late Show just with his

webcam moving his arms. Stephen moves his arms. It's nuts, right, And imagine that kind of technology in the hands of artists. Start thinking about them as not like something that's gonna destroy our creativity or gonna replace writers an artists or whatever. This is gonna be a typewriter, This is gonna be a paintbrush, and people will start building and using it to understand their own creativity in a new way. Of course, as always, it's up to us to make sure we

use these new tools for good. If I build a shovel, okay, and you decide to go to the beach and digging sand, you're biased. You're digging in sand and guess what You're shoveled only to an up sand. The same as true for AI. If you give it a certain type of data set, I can promise you whatever I get out reflects the data you're put in. It's up to us,

the people, to make responsible decisions. And as we want to create equal opportunity and evadicate certain biases society that exists today, is up to us to do it, and I promise you, if you work hard on this, technologies will reflect that. But even Sebastian, one of Silicon Valleys Great Optimists, recognizes the risks all technologies can harm people. In fact, technologies can be abused to harm people, Like my kitchen knife, which serves me a great purpose every

time I've guests over in shopping. My produce can also be abused to harm people. In the next episode of Sleepwalkers, we dive deep into the ability of algorithms to cause harm. We traveled from China and the social credit system to a parole board in New York and we speak with people building technology they believe will make us safer. I'm as Vlachen, see you next time. Sleepwalkers is a production

of our heart Radio and unusual productions. There's so much we don't have time for in our episodes, but that we'd love to share with you. So for the latest AI news, live interviews, and behind the scenes footage, find us on Instagram, at Sleepwalker's podcast or at sleepwalkers podcast dot com. Special thanks on this episode to paw Suris, who introduced us to Oscar Ross and Benjamin and to artificial intelligence, which composed over half of music in this episode.

Could you tell which was which Sleepwalkers is hosted by me Ozveloshin and co hosted by me Kara Price. Were produced by Julian Weller with help from Jacopo Penzo and Taylor Chacog. Mixing by Tristan McNeil and Julian Weller. Our story editor is Matthew Riddle. Recording assistance. This episode from Joanne de Luna. Sleepwalkers is executive produced by me Ozveloshin

and Mangesh hatt Together. For more podcasts from my heart Radio, visit the i heart Radio app, Apple Podcasts, or wherever you listen to your favorite shows,