Robin Sloan: Full Interview

Hey Siri, play Machine Yearning.

Here's Forgive and Forget, Isabella Machine Summer's remix by The Kooks.

Machine Yearning podcast.

Hmm, I'm not finding anything for that.

I'll do it myself.

M-A-C-H-I-N-E-Y-M-A-C-H-I-N-E-Y-M-A-C-H-I-N-E-Y.

It's Machine Learning from Assist. It's a podcast where we think and dream about the future of AI, the talking internet, and how we're reshaping our culture. Today it's sci-fi writer, social observer, and dedicated AI enthusiast Robin Sloan. Robin's path into AI has been fascinating. Starting with his neighborhood library in the Detroit suburbs, weaving through the early days of Max, and most recently, Robin has been intuiting and bootstrapping his way into some fascinating co-writing experiments with the machine. We find inspiration in the humility and daring he brings to his writing and the collaborative process. Robin has been talking about how, with his machine learning co-writing experiments, quote, the goal is not to make writing easier, but to make it harder. Robin is funny, welcoming, And he challenged us to think about new ways to look at what's possible when the machine delivers lovely text like, "The slow sweeping tug moved across the Emerald Harbor." We loved our time with him and we think you will too. Enjoy.

Hey man, how are you?

I'm good.

Thanks for coming by.

My pleasure.

I'm going to start off, I want one question. Did you always want to be a writer?

Yeah, yeah, yeah. I was a public library kid, you know.

Really?

Yeah, absolutely. Spent a lot of time in the— Troy Public Library in Troy, Michigan, where I grew up. And on some level, I was the kind of kid that, like, as soon as I saw those books on the shelf, I knew I wanted to have a book with me.

Do you remember the first book you saw?

The first book? No. I do remember I was a devotee of the Choose Your Own Adventure books. There was a little wire spinner rack that had those on it. I mean, I like books of myths too, big picture books of, like, Greek myths.

Why?

Oh man, I don't know. They just were They were magnetic. One in particular, there's actually a whole cult around this book. It's called Dolère's Book of Greek Myths, and it's huge. It's like the size of a coffee table book, but it's for kids. Beautiful illustrations, just these total, like, rich, colorful illustrations you can, like, fall into. And these really great, crisp, like, compelling kind of renditions of all the Greek myths. And I used to check that out from the school library, from the elementary school library. Every week I would just go back and like check it out again. Basically, I had a lock on Greek myths. If there was any other elementary school student in my school who wanted to like learn these Greek myths, that was not available to them.

Seriously?

Yeah, for a year.

When did you go from reading to writing?

You know, pretty immediately. And that's where it ties though, from the beginning, or almost the beginning, into computers, because we had a Mac Plus in my house.

What was a Mac Plus?

The Mac Plus, that was like—

my brother got me into Macs when they turned colors.

Yeah, yeah. Oh yeah, this is pre-color. This is, this is black and white screen. This is somewhere in the middle of the '80s, maybe '86, '87. My dad brought home a Mac Plus. It was really the first computer that I had used. There probably had been some Apple computers at school or whatever, but it was the reading and the writing and the books plus that computer and like learning how to use that computer and kind of hack on that computer a little bit. A lot of my earliest creations weren't like handwritten on paper. They were typed into whatever text editor that was.

Well, that ties into now. It seems like you've always been fascinated by these writing tools. Yeah, like, talk about these writing tools you've been building now. I mean, do they feel the same as then? You're like— your blog post I read, you're basically in a DOS text editor.

Yeah, yeah, it does, you know, it does feel the same because there's that sense of like— I'm probably a little too interested in the tools sometimes, to tell you the truth, you know, to my own distraction. I'll get kind of wrapped up in thinking about the tools and trying new tools and creating my own tools where probably I should be spending at least some of that time actually using them and, and creating more things. But I don't know, that's just my brain. And yeah, and so given any new technology, and I mean, this has been true of like the web, the early web, it's been true of mobile phones and like the kind of the interface of the app on the screen, and now machine learning stuff. For some reason, my impulse is always like, Yeah, how can I learn to write something with that or on that in like a new way?

And when did, when did it go from a new tool to like kind of the thing— I'd love you to explain like what I've read about now and what you're working on now as it relates to machine learning. I almost said machine learning. Point taken. And what got you inspired by these new tools? Where did, when did it go from writing to you saw machine learning to applying that?

Yeah. I had, like everybody, read a ton about machine learning and AI tools and all the kind of hype and self-driving cars and all that kind of stuff for years. I ran and thought it was interesting, but no more interesting than any number of other kind of like cool high-tech things that you read about in Wired or The Economist or The Atlantic or whatever. I ran across a blog post that a lot of people read and liked by this researcher named Andrej Karpathy. What does he do? You know, now I think he works at Tesla. He, at the time that he wrote the post, he was at Stanford. He was a PhD student at Stanford and had been doing all sorts of work, you know, on just all kinds of different neural networks and applications. And on top of all that, just had, and I think still has, a real gift for explaining it and kind of sharing what he knows. So he'd written this post about using recurrent neural networks to generate text. And I'm pretty sure at the time I read the post, I wouldn't even have recognized the term recurrent neural network. And I didn't. And again, it was my understanding of all this stuff was very kind of superficial. So the post was interesting. His voice is just really engaging and welcoming, you know, it's not sort of the voice of the engineer who's like, well, you probably won't understand this, but we'll try anyway. He was like, all right, let's, I'm going to break this down and tell you how this works.

I love how you start moving the water bottles around like they're Legos.

That's right. You could see the engineer versus— yeah, yeah, yeah, yeah. The, um, the thing, the thing that really caught me though, it wasn't the technical parts of the post or the neural network stuff. It was that he, of course, shared examples of text that these systems had generated, and the text was really weird and really beautiful. I mean, that's what it was. I saw there, like, in the shape of the text itself, something that I found really artful and kind of magnetic and interesting. If that hadn't been the case, if it had been technically interesting, but then, you know, the text it had generated had been really flat or kind of generic or blah, then I think I would have moved on to the next post and the next post and the next and never even have spent another second thinking about these tools.

Do you remember what it was that was weird and beautiful?

I don't remember from, from his post. I do remember there was a phrase, a terrific phrase that leapt out at me after I had started tinkering with the tools myself. And this is probably the next week. I mean, pretty quickly after that. Yeah, I was like, I should, I should try that out and see if I can, you know, get anything interesting come out. There were a bunch of tools all kind of bundled up that were just super easy to download for a kind of enthusiast-grade programmer like me. I could just kind of download it, get a bundle of text, set it up, get it training, and within a couple days start to see some output. And there was just this one little phrase. It's not going to sound very magical, but it wrote something. It was kind of this big blob of text because it kind of goes on and on and on. Of course, these machines, they do the text so well, but on the level of the word and the phrase, it's so beautiful. But then you zoom out a little bit and you get to a level like the sentence and the paragraph, and it's basically like an insane person writing. Because at least at the moment we're in now, they don't kind of understand these larger structures of like the idea of a story having a plot or where it's going. So who knows what else was happening in this text, but I remember it wrote, "The slow sweeping tug moved across the Emerald Harbor." And it's a tiny thing, but like, "the slow sweeping tug," that's just a cool thing to say and a cool thing to read. Like, "slow sweeping" is kind of a weird word. Yeah, absolutely, exactly.

I don't know, some book. I don't even know who would have wrote it. Moby Dick? It's like somebody wrote this. Totally, totally.

And that's saying a lot. That's saying a lot that it was this thing generated by this computer program that had whatever that is, that touch of art or beauty, the unexpected. So that in my own— that was maybe the second like stroke of the piston. First I read this blog post, then I tried this stuff myself, and I read about the slow sweeping tug moving across the Emerald Harbor. I'm like, I will pursue this.

Is the machine the artist thinking? Artistic thinking, or is it the person?

Well, it's a combination in that case, because of course, you know, in this particular instance, you're training it on text, and so there is an underlying art, which is all the art that's, you know, gone into all the books or, you know, screenplays or novels or whatever that you're kind of feeding into the system. So that's all there, but boy, there's a lot of it, and it's not like— I mean, sometimes you can do this very specific thing where you train a neural network on Shakespeare and it like learns Shakespeare. That's That's fine. All the experiments I've done have been much larger sort of bodies of work. So it's not just one author, one voice, it's tons and tons and tons. So it's all in there getting kind of just chopped up and stewed together. And so I actually think then when you look at the text that comes out the other end, I actually think it is appropriate to ascribe some of that, some of that beauty and some of that creativity to the, to the machine itself and to the neural network. Yeah, I think it's doing something new in the same way, in the same way that Frankly, I think my brain does a certain kind of work when I sit and read like hundreds and hundreds of books and then go to write something myself. I mean, I think the two processes are— it's not the same, very different levels of complexity and sophistication, but I think they're not totally different either.

What did you mean when you said the goal to everything you're doing is not to make writing easier, it's to make it harder?

Oh yeah, yeah, yeah. I wrote that in one of the posts that was explaining—

I read it and I was like, yeah, it was like one of those lines where it It was like, wow, that makes a ton of sense, but I have no idea what you mean.

Yeah, yeah, yeah. You know, it's not to kind of retreat into analogy, but I think of other tools and other domains, like something like oil paints. Have you ever tried to paint with oil paint? I've only tried a couple times.

Oil painting, like, my house at home, and I couldn't get it off my hands for weeks because I'm like, why are we using oil?

Yeah, well, so this— it's a really gnarly media, it turns out. It's really hard to work with. Nobody would— nobody ever said, "Oh, you know, I want to just paint a nice easy— I just want to have a chill afternoon of painting or of image making. Oh, I know, I'll use oil paints." They're really challenging and it takes a lot of skill to do all those effects that we now admire, you know, on the walls of museums. People who play games sometimes talk about a skill ceiling, like the people who are really into like really challenging video games. They like them because the difference between people who are competent and people who are really, really good is like quite a difference. Whereas there's other games where you just get good at it and that's all there is. You're good at it. There's no such thing as being very good or extremely good or world-class good. And I think the same goes for artistic tools. Like, you know, if writing in a text editor is like drawing a portrait with a pencil or a piece of charcoal, What is writing with oil paints? What is that thing that's actually more difficult and it kind of like you have to learn some new techniques, you have to learn how to handle materials, but then in return maybe you can produce something that's even more beautiful. Machine learning.

Since we've been talking about writing as a craft, the challenge to improve your craft was defined in a few ways. Now, as Robin Sloan is describing, Machine learning, the tools and the ways you have to stretch as a writer and collaborator, have changed the game. Even if collaboration is with the machine. This is the essence of this podcast.

Machine learning.

Machine learning is all about the ways we navigate a new world where identity and technology and commerce and security have to live together in ways we can't even fully imagine yet. We have some great guests lined up, but we want to hear from you. Who would you like to hear on machine learning? Let us know. DM us @Assist on Twitter. Coming up, Robin walks us through the importance of patterns and structures and how a machine that doesn't even understand what a word is can still learn how to communicate.

I want to know more what the tool is.

Sure.

Because I don't understand. I mean, I guess to take the listener there and me there.

Yeah, yeah, yeah.

I want to know what is this tool?

Yeah, yeah, so basically it's— and this is where, you know, it's gonna kind of expose the limits of my own understanding because I'm not one of these machine learning researchers, but I have used the tools a lot. It's almost always, not 100% all the time, but almost always something called a recurrent neural network. And it's actually, at its heart, it's kind of beautifully simple. It's a neural network that learns about patterns, about sequences, and it could do that for musical notes. It could do it for numbers and some weird cryptic series that you wanted to understand, like maybe you're trying to model a weather pattern or something like that. It can definitely do it for text, right? Because that's just text, is always some kind of sequence, a sequence of words or sequence of characters. And really all it learns, I mean, you could, you can imagine if we were just doing something simple on pencil and paper trying to come up with a model of like English, we would determine really quickly that given the letter T followed by the letter H, it's very likely that the next letter is going to be E or A, very unlikely, like vanishingly unlikely, that the next letter is going to be Z or Q, right? So you imagine that as a little kind of like module of statistical understanding.

But the engine doesn't understand English.

That, I mean, that's actually the beautiful thing. It doesn't understand English. On moment one of like training this recurrent neural network, it's not like you've coded saying, okay, get ready, You're gonna break this up into words, and here's what a sentence is.

It doesn't even know what a word is.

Oh no, no idea. It is just like opaque data.

Does it see the T as a line that's horizontal and a line that's vertical? Yeah, you know, see the T as a letter?

It sees the T as a letter. It sees the T actually— I mean, usually the way it sees it is as a number, as, or as one in an array of, you know, 128 potential values or 256. It's really interesting, and that's, that's the astonishing thing. Starting from that kind of really, really cold start of just having no preconceptions about language or the world and then training itself on— again, and this is not like one short story or even one novel. I mean, to generate, kind of to bootstrap this deep understanding of the structure of English and the way that people write. The experiments I've done that have, to me, been the most successful in terms of generating the text that's been the most interesting and like the most interesting to read have been all with a big body of text that I collected from public domain scans of old science fiction magazines. And so it's hundreds and hundreds of megabytes of old science fiction stories from like the '60s and the '70s all kind of glommed together. Lots of different writers, different eras, but all with that, you know, that set of subjects and kind of that tone, that science fiction tone.

So quiet. And then you use that how? Like you're sitting there, you can hear you're typing your fingers on words and let— or you're typing your fingers on letters.

Yeah.

And then you have this robot that sits in your head, like, and you hit robot, help me. Like what happens?

Well, so that's, and for me, that was the second step. It initially, 'cause of course we're just using these tools that kind of researchers have developed, it's all happening on the command line. And I didn't actually see how I could write with that. It was kind of a curiosity. It'd be like, whoa, look, the computer just generated like, 2 pages of really weird, interesting text. Okay, I'm not really sure like exactly how to use that. And so I did— it was— I was riding a BART train actually coming into San Francisco. This is maybe a year and a half ago now, almost 2 years ago, riding a BART train coming into San Francisco thinking about this stuff because I just kind of started tinkering with it. And I had the notion that what I wanted was really autocomplete. I wanted a super creative autocomplete. I wanted to be able to start typing something myself hit Tab or some key and have this system, this statistical model of, you know, in this case science fiction, complete the thought for me. And so that's what I ended up building. Interesting. Yeah, and so, you know, you could type, uh, the ship landed on the planet's surface. Captain Astro stepped down and saw— and then you're like, well, what did she see?

I don't know.

Let's see what, let's see what the computer thinks. Hit Tab and it'll fill it in. And it, I mean, it, and it could say anything. What's good though, what's cool is that it's not garble.

So how are you thinking about this phenomenon? How should I think about this? What, what is like, where is this going?

Well, you know, I think there's kind of two branches for these kind of technologies. And when I say these kind of technologies, I guess I mean programs that have an increasingly sophisticated understanding of how we write and how text comes together and can kind of fill in the blanks in interesting ways, right? And I think, I mean, obviously autocomplete exists and is widely used everywhere. So that's—

fucking ruins my text every day.

That's the— I'm totally— I text sci-fi a lot, like a lot, at least a couple times a day. And for some reason autocomplete has not figured out that I want to say sci-fi. It always says sci-do. Which is not a thing. I don't understand it. Said you. So that exists, obviously, and it proves that, like, these are gonna— these tools are gonna be an important part of our life— lives. I think there's two branches now going forward. One is, to me, a little darker and more dystopian. The other one is more interesting. The darker, more dystopian one is a future version of Gmail where it pretty quickly understands what kind of email you're writing and basically offers to write the majority of it for you, or to like make a pretty good draft of this email for you to then go back and edit and then send.

That's dystopian to you.

Yeah, it is. Oh yeah, it is. It really is. Just only because— only because I really— maybe this is just—

no, I get it.

You know, but it's because I think it's because of just who I am and how I make my living now. The idea of words that are essentially signed by a person, words that that pretend to be from another human, but they're not really. They're essentially like a form letter generated by a computer. I just think that's gross. I just think it's kind of an insult to inflict that on another human being.

When does your dystopian lapse and you're like, "Fuck it, it's just— that's just how it is," or does it?

Well, you know, it— that's a good question. I don't know. I have tended to focus on the way, in a way, almost heading off the dystopia. And this is what I've done. Like, in the tools that I've been building, I've been trying to sort of produce and share implementations of these techniques that are designed not to, like, make the world more efficient and, you know, make communication faster and easier, because I frankly don't care about any of those things, but to make stuff weird.

To—

make stuff weird.

Yeah, to, like, to kind of weird it up, to add a little perturbation of strangeness and, like, whoa, what? Did I just read? Or like, what is going on over there?

Give me an example.

I mean, the example is in—

Give me some weird shit.

I want some weird shit. I should have brought a short story to read. I mean, the answer, the weird shit, is in sort of being in conversation with these machines. And in particular, machines— this is the thing I think a lot of machine learning people in both research and industry overlook. They focus on the engine, assuming that the thing you train it on is going to be something big, broad, generic, useful. I mean, imagine all the emails stored in Gmail, or imagine just tons and tons of like news articles from Reuters or the New York Times. This is the kind of stuff that people use. As we discussed, the things that I've tended to use have been old sci-fi stories. I've done experiments where I've taken took two different bodies of text, really different, like one kind of horror fiction and another the text, the transcripts of Supreme Court proceedings, and mashed those together and told the system, like, all right, learn how to generate text that fits both of those at once. Yeah. And the goal—

Did you take, like, Republican and Democrat?

Absolutely. I mean—

Like press releases.

The truth is there's a ton of interesting work to be done in thinking about what we train these systems on. And I think that's gonna have to end up being the, the kind of domain or the work of a lot of writers and artists and weird people, because I just don't think Google or like Stanford grad students are ever gonna do it. They're just kind of focused on different things.

The efficiency game is what it's used for today.

Yeah, that's right.

You're focusing on something that's, I think, way bigger, but everyone sees it as like an artistic kind of thing. But it feels like everyone in the human side of things— we're coming out of 100 years industrialization. We're like, okay, those are probably gonna be lost jobs, maybe. And if so, then the only thing that matters is creativity and creative thinking. If the machine becomes creative, what happens?

You know, it's interesting, just— I mean, that's a big question, a big thing. It's a big thing to think about. Even to just narrow it down and think about text, you know, what if What if I wasn't working on tools to kind of augment my own writing or the writing of a person? What if I really was making a machine that could write a short story? And that's not implausible. I mean, there's the pieces of the puzzle are sort of there, and you could imagine with a few more steps forward, just a few kind of increments, I could imagine a system where I say not just finish this sentence, but okay computer, write me a short story, and it writes one and it's actually Pretty good.

Have you heard of the infinite monkey theorem?

Yes. Yes.

I just heard about this two weeks ago. Yeah, that's Robert.

Yeah.

Yeah.

Yeah.

He's like, here's the deal, right? If you have infinite monkeys and you give them a keyboard or typewriter and you let them just type, one of them will write a Shakespeare novel.

Yeah. Yeah. Yeah. I mean, I'm like, what? Yeah. I mean, what it actually, in a way, it loops back around because what What that leaves out is that you need to identify it somehow. Somebody would have to be sitting there reading the output. This sounds like a great job, by the way. Somebody's gonna get stuck with this job reading the garbled output of the infinite monkeys until they're like, "Yes, here it is. Here's Shakespeare." Um, at which point you're like, maybe we should have just had that person write the Shakespeare in the first place or whatever. And I think, but I think that's important. Um, it turns out the people who are to me smartest about this stuff, and it's not just the sort of the text-based, you know, machine learning experiments. There's a lot of people doing work with images, people doing work with sound and music. And all the people who seem to me smartest and most grounded about these tools, they insist that there is this super important step at the end, which is a kind of filtering or editing or curation. You essentially ask these systems to produce a bunch of material. It's really weird, interesting material, but like that's not the end of the process. You as an editor, as an artist, as a thinker, as an editor have to sit there and say, "Mm, this, not that. We'll use this, that we'll throw in the trash can." But it's, I guess I would say that's one of my like sci-fi predictions I feel most confident about.

What? Can you make that prediction very clear?

Yeah, in some number of years, I don't know if it's 5, I don't know if it's 10, most text editors that people use in email, in their offices, are going to have totally augmented writing capabilities. And you're just tweaking it. Yeah, exactly. You're the editor. You're kind of building a document block by block, and then, and then yeah, going back and making some changes or saying, oh, that doesn't sound like me, or— but then again, maybe you just select a whole paragraph and press the key that means like make more friendly and it adds more exclamation marks and maybe a few emojis. Yeah, yeah, yeah. I think that's, I think that's, it's coming for sure.

And now my email marketing software, my blog software, my chatbot software for messaging, my Alexa software, whatever is powering our language, it's suggesting what it knows I already should do. Yeah, yeah, probably tested it too. And I'm just like I'm more of like the legal department approving it.

Yeah, and I mean, in a weird way, but then, and it raises the interesting questions because of course, you know, humans, we always, we like get these tools and then the tool, the tools change us and they change our behavior. It raises the question like, then will there be this little thing where you'll write this email to your colleague using this robotic system, but then you'll put some little thing at the bottom that's the little wink of life, you know, there'll be some way of saying Hey dude, or a little emoticon or some like secret code word that becomes evidence. Like, it's because it's something that the robot could never write. I don't know what that would be, but it does— I mean, if it's all a perfect masquerade, then some people are gonna find a way to poke through or a way to undermine that system. It's gonna get weird. I think it's— it already seems very weird to me in the era of like Slack and everything else to work in a a very tech kind of saturated office. I think it's just gonna get weirder and weirder.

Fascinating. I was at this talk one night. It was people from IBM Watson, and they were showing how they can see cancer in the photos. They can— the machine can learn if it's cancerous. They claim it was like 99% effective. The human error rate for detecting cancer was like 80%. Yet the human was required to make the decision at the end because we still trusted the human who had a worse error rate. Yeah, and then they got this huge argument. One, I think the Watson guy was full of shit, totally full of shit. But two, that's why I think you needed the human. Yeah, the other guy in the room was like a real scientist, and he was like, if you believed in it and the percentages are right, you're gonna kill 19% of people because you don't believe in the machine. And I was like, fuck, I like blew my mind. I was like, I don't know what I would do here.

There's—

I—

boy, we are gonna be faced with more and more of those challenges. It is—

whose fault is it?

It's deeply uncomfortable. Yeah. Well, I mean, I actually— I think one of the reasons— this is pretty gnarly— but actually one of the reasons that humans stay in the loop in a lot of these cases is that it's not clear. If there's not a human, it's not clear whose fault it is. Or worse, for a big company like IBM or Boeing or whoever, Tesla, the people who make this autonomous software, if there's not a human in the loop, it might mean that the fault is theirs. And like, they don't want that.

Who's the insurance go after?

Exactly. Exactly.

Do you think that's real? Like, that's a huge—

I do think it's real. Yeah, there's a, there's a phrase kind of bouncing around a little bit in kind of AI policy circles. Um, the idea of a human operator or a human, a little, little bit of human judgment as a kind of a moral crumple zone, you know, the person's the one who can absorb the shock of something going wrong. And it's not that we would never figure that out. There's going to be new policies and new precedents set around liability in terms of AI decisions. But it's super unsettled right now. And I actually think it is just easier for a lot of, for a lot of these companies and institutions to be like, yeah, it's, it's Robin. He was the one driving, or he was the one who like looked at that screen and pressed a button that said, Okay, I approve.

The limiter to advancements in the technology is because I need someone to blame.

Yeah. I mean, that's gonna change. That's gonna change, but it's such early days now. Nobody knows anything.

My question to you would be, take— that's like a very dark version.

Yeah, yeah, right.

But take it in a world that's, you know, very politically correct today. We're politically correct, right? It's really hard to not be politically correct, although some people aren't, and then the internet goes crazy. Do we care if robots are politically correct? Do we care if the sentence it produces sounds racist or terrible?

Yeah, I think we do. I think we do deeply. There's two pieces to that. One is, of course, they only learn what they're taught, and we've seen this time and time again. Yeah. Training sets have— of course, you know, of course they have biases built in. They're human artifacts. They don't just kind of come from the cosmos. They are assembled, put together, and fed into the computers by people. So, so to the degree that what comes out the other end of these systems is biased or weird or like privileges one group of people or one set of ideas over another, I mean, that's actually telling— just telling you something about our world. It's not telling you too much about the machine itself. But the other thing I'd add is that You know, I think it's like using a machine does not free anyone, a company, an entrepreneur, or an artist, from these obligations to think about being like a responsible actor and speaker in the world. And so for instance, in all the stuff that I build, at the very final layer, I put in a word filter. Like none of these machines are ever that I create, even though, I mean, because I don't know what's in that Training set. That's 150 megabytes of text. I didn't read it.

You don't have people to read your monkey.

I didn't. Yeah, exactly. Exactly right. Exactly right. The monkeys, the monkeys are on the loose. And these are, these are some old sci-fi stories too, which is, it was, it was a different time. And they're actually, there, there is some, some—

It's like watching ads from the '70s. Like it's not appropriate.

Absolutely. Absolutely. Um, and, and knowing that and, and I just being kind of mortified by the idea that someone trying one of these tools at home, kind of playing around with it would type something and then hit the tab key and then see something that was like really bad, like sexist or racist or just ugly in any of a number of ways. I just found that so— to my knowledge it hasn't happened, but just the vision of it is so mortifying. So yeah, I built in some safeguards to make sure that that won't happen.

Do you think that the safeguards are one of the most important things to be built in this space?

Yeah, I do. I do. Now that you've done it. I think it's really important. Yeah. And it's just in part because they just prevent weird stuff from emerging, weird, damaging, dangerous stuff, and also because thinking about them is a really good habit, you know. It just— so much of it, so much of the work tends to be focused on making it work, and you get— and when it just— when you see that spark of like, "Oh my god, it's writing a sentence!" Yeah, it's so exciting. And but— and so, so thinking about the safeguards makes you just kind of take those next few steps into like, Okay, it works. Good. How is this gonna work in the world? Like, how is it gonna fit into a broader world, not just my little laboratory here? And I think that's important.

From the limitations of potentially politically incorrect vintage sci-fi to the 10 million monkey model, AI proves that GIGO—garbage in, garbage out—hasn't lost its power.

Machine learning. Machine learning.

We hope you'll take the time to visit Robin at his digital home, where you can find his blog, get your hands on his books and other writing, and sign up for his newsletter, which is really like getting a message from a friend. You can find him at Robin Sloan, and Sloan is spelled S-L-O-A-N. Coming up in this final section of today's Machine Yearning, Shane and Robin dig into Robin's experiments with using machine learning to write music and how playing Go against the machine taught the grandmasters that there were levels to the game they had not even considered. In a conversation that was already really fun and thought-provoking, this section illustrates the degree to which human-machine collaboration might really be a 1 1 5 reality. So let's get to it.

I also want to talk about, because you're using machine learning for music stuff. Yeah, a little bit. Yeah, like, tell me about it. Where— when did— when did the music stuff start? And then how did you— what are you doing with machine learning and music?

Yeah, that was one of those funny kind of curveball things that you just could never predict ahead of time. I'm not a musician myself, or even super deeply engaged by like music production or anything like that, but I am a writer, a fiction writer, and they do make audiobooks out of basically all novels these days. It's a huge market. It's like the part of publishing that's growing the fastest, and it is actually growing a lot. Really? So what it means is the audiobook producers at the publishers are quite emboldened, and rightly so. And so for both books, they reached out to me and said, "Well, okay, we're getting ready to record this audiobook. Could you write something new to kind of slip into the audiobook, or like, can you think of anything else that would make this just a little different, a little special?" 'cause I wanna find a way to kind of differentiate it from all the other audiobooks out there. And my new book, a part of the plot actually hinges on this kind of mysterious form of music that I describe in the text. And so it seemed actually, I don't know, obvious that in the audiobook you'd actually get to hear a little bit of this music. Maybe even frustrating if you got through the whole audiobook. I spent a lot of time in this book just—

The music from the book. From the book. This is sourdough?

This is sourdough. Yeah, that's right. And yeah, I actually had the thought because we're talking about the audiobook and I was like, man, if you get to the end of this book and you've heard me describe this mysterious music in like a language that no one could quite understand 10 different times and you never actually get to hear it, you might actually feel a little bit cheated. So I decided it would be a good opportunity to try to apply these tools to something really concrete. It felt like something I could bite off. It wasn't trying to produce like an album of generative machine learning created music. It was just— that's happening though. That is, that absolutely someone's— more than one, I'm sure. There's a whole host of people hard at work in their studios trying to be the first person out of the door to do that and actually make it popular. No, this was great because it was just these little snippets. It had to be these little bits that we could kind of fit into the text in different places. So it was actually ended up being a really similar process to the text, except that instead of text, it was audio waveforms.

And how do you make it? What are you using? What tool?

How— so, yeah, actually doing— so in this case, it was a piece of code. It was, boy, one of those great internet open source stories. There had been a paper describing a system called Sample RNN, and again, the RNN here is for recurrent neural network. So you see, it's all connected. It's all this sort of same underlying core tool. The paper had described these great results with sound. The people who put it out, had put out an implementation. I couldn't quite get their implementation to work. Someone else, a like renegade genius in the UK named Richard Assar, did an implementation of his own, you know, just as one does. He was like, "I am interested in these techniques and I would like to write a version of that in my language of choice." So he writes it, puts it up on GitHub. I find it because I'm looking around for a tool that I can use to do this audiobook thing. Get it up and running on my computer, feed in the audio. In this case, it was this sort of Croatian choral singing that I've just loved for a long time and was actually— it's what was in my head when I was writing my descriptions of this fictional music. So there's a kind of a parallel path here. Like, I had used it as my inspiration in a sense for what I put on the page, and then I fed it into this recurrent neural network, and it worked great. I mean, it was actually spooky.

I want to finish with one thing you said earlier, and you talked about the dystopian type future, but you never talked about the other side. What do you think the other side is that is actually a great thing that can happen?

I think we've actually gotten a glimmer of the— not utopian, but, but really exciting kind of machine learning future with the AlphaGo project. That— what is that? DeepMind wrote a computer program based on neural— yeah, it's a, it's a a wholly owned subsidiary of Alphabet, and they were the first people to write a Go program that could beat the best Go players in the world. And Go is a game. Go is a game.

Like chess? Like, what is Go?

It's, it's, um, it is like chess in its— I mean, in terms of how seriously some people take it, it's definitely like chess. It is a more ancient game. It's really simple. You recognize it, I think. It has these black and white stones on a big grid. It's called checkers, right? This is like super Super checkers. It's a hard— it's a really hard game. A lot of strategic depth. The people who are good at Go are really, really, really good at Go. And there had been a lot of debate about how long it would take for a computer program to beat the best Go players. So DeepMind did it. So that's just, you know, interesting and kind of a milestone. But what's heartening and to me really, really exciting is actually what happened next. Because these Go players, like the super grandmasters of the world, some of whom got beat by AlphaGo, the program, and others who just kind of watched the matches, their reaction was, "Oh my god, we didn't know you could play Go like that." Like, a few even described it as being somehow analogous to like playing Go against an alien. Like, if aliens landed on Earth, they're like, "We are here not to conquer." We want to play your games. And that's amazing that it's not just— it's not actually just a story of like clash and conquest and like, oh, too bad humans, looks like you're not the top of the food chain anymore. But instead, all these people who spent such— so much time studying this game, loving this game, in fact now realizing there's— there are whole new avenues that they never even contemplated and in a way the door was opened by this computer program. Could the same thing happen for music, for writing, for—

so when you said— when you said it's not about making writing easier, it's about making it harder, it makes it harder because it actually makes us better.

I mean, that's the hope. And, and you know, it comes to creativity, it's probably less the case that there is better and worse and more just there's just more. I mean, there's, there's more ways to write, there's more things to learn, more ways to paint, more ways to make music. And so that's, that is the optimistic vision, that just like any other creative tool, these machine learning tools make more things possible, and, and people get really good at them.

Wow. Yeah, it's interesting you say that to end this, of like, I never thought about it that way, where What it's actually giving us is a— is a— because it's not a human, it doesn't have the emotion to get mad at, but it gives me another view and of openness that I didn't have before that hopefully makes me more open.

Yeah, I mean, I can report it's, it's easy to kind of talk in big sweeping generalities and kind of get into pretty heady philosophy about this stuff. I can report from the ground level, working with these tools myself, they've changed the way that I think about my own writing. Wow. They've produced, on the most basic level, they've produced bits of language that I think are really cool and interesting and that are making their way into my writing, you know? My next novel is gonna have little bits of this machine learning generated text in it.

Well, I thank you for coming by today.

Hey, my pleasure. It was fun. Bye.

All right, thanks for listening. Get in touch on Twitter @Assist. DMs are open. We're super interested to hear who you think should appear on the podcast. Machine learning. Please subscribe so you never miss an episode and share this with someone who cares about how we make sense of these changing times. Machine Learning by Assist is made by Paul Ciuffo and Michael Alsesser for Limina House. Have a great day.