POKEY THE PENGUIN
3rd wave crusade barrelfishing (quendinburg)
- Joined
- Feb 17, 2022
- Messages
- 119
- Reaction score
- 388
Author's note: I gave this talk in February 2014 as a guest of honor at Seattle's Foolscap science fiction convention. My other presentation at Foolscap was a computer art installation featuring speculative-fiction versions of my creations Ebooks Brillhantes and Hapax Hegemon; these are the monitors mentioned at the end of the talk.
Constellation Games, quoted and used as an example text several times, is my first novel. It talks a lot about alien and machine intelligence.
I slightly edited and modified this talk for the Web, including adding a lot of hyperlinks.
If you habitually talk to your phone, think about the sort of things you say to it. Would you talk to another human like that? What would happen if you asked your phone about itself? You can try it. You'll probably get a canned, jokey response that says nothing. The sort of response a well-trained butler would give.
So I would prefer not to anthropomorphize pieces of software. But I can't stop, because I'm a fiction writer. I anthropomorphize everything. I have to anthropomorphize a fleeting idea in my head, and then I have to write down, like, ten or fifteen words of description, and I have to convince you that behind those words there is an independent intelligence with a rich inner life.
Besides which, I'm a science fiction writer, and if I can anthropomorphize pieces of software, I've got some characters.
So I've been trying out the idea of thinking of pieces of software as alien beings of various types. It's a fun character, almost a stock character. They're different from us. They understand English to varying degrees. None of them really understand humanity, but that doesn't necessarily mean they're stupid. Maybe we have something to learn from each other.
The second front in the AI project is not an official research project, but it has also been a big success. This is the project to make humans more computer-legible. To train us to act in ways that can be processed by these weird, obsessive aliens that don't really understand humanity.
When we call the bank, an alien answers the phone. We communicate with the alien by pushing buttons on our phone, or by shouting words one at a time, as if we were trying to get a French person to understand French. And if it doesn't work, we demand to talk to a human. But when we do, we talk to a human who's been trained with a script to act only in certain, pre-programmed ways. The human is being paid to pretend to be an alien who speaks English really well.
That's not so bad, because the aliens don't read. They don't care about the content of my fiction. I can write whatever I want, so long as I think I can convince a human to publish it and a bunch of other humans to read it.
You can think of Twitter as an alien library that can contain an infinite number of texts. Due to a quirk of its biology, it cannot metabolize any text longer than 140 characters.
When I write something for Twitter I identify myself and I type into a text box. Whatever I typed is stored in the library. A lot of incidental information is stored alongside it: the current time, the fact that it was me who wrote this, and my physical coordinates.
This is Twitter's food. It can only accept information that fits this specific format. If I want to feed it something that's 141 characters long, I can't do it. That last character has to go into a separate piece of text, encrusted with its own set of metadata. This is a biological limitation. Twitter cannot contain a 141-character text any more than I can fly.
This biological limitation changes the way I write. I use abbreviations I would normally never use. I reword things to make them shorter, even when that makes them sound stilted. Sometimes I can't get it to work, so I give up and don't write anything after all.
What we see now is a kind of symbiosis, where the aliens make maps and remember things for us, and in return we only do things that the aliens can understand. And of course we fix the aliens when they break.
Okay, I love aliens. I create aliens all the time. Real aliens made of software, like Twitter, and fictional aliens made of words, like the ones in my novels. And fictional pieces of software created by fictional aliens. But this thing that knows a lot about our relationships yet doesn't understand English, doesn't sound like a trustworthy alien. At the very least, it sounds like an alien that can be easily tricked into doing a lot of damage. And we've discovered that frequently, in the shadows, there lurk sinister humans—the secret masters of the aliens.
In computer science we call this a data structure: specifically, it's a list of trees. You are experiencing my talk as freeform English prose, but I've also unwittingly structured it for consumption by a machine. A machine that probably doesn't understand the sentence "This talk has slides", but which now knows that "This talk has slides", whatever that is, is a more important concept to my talk than "(A list of trees)".
Okay, but.
This is why we build aliens like Twitter. There is something appealing about constraints. We thought "maybe if we only had 140 characters, we'd be forced to make every one count." On the same logic that if you knew you only had a month to life, you'd get off the couch.
In real live, the 140-character constraint didn't solve any of the problems I just mentioned. If you want to say things that will hurt people you don't know, Twitter is probably the best medium ever invented. But this oppressive, fascist-seeming constraint has led to a breakthrough in human creativity and in human-alien relations.
How is this possible? We're American science fiction fans! We hate constraints! And it's true that constraints that are forced on you are terrible. But voluntarily-chosen constraints are the source of creativity.
You may only paint straight lines. You may only use the primary colors: red, yellow, and blue, plus black, white, and grey.
You may only paint rectangular shapes in solid colors. A single canvas may contain at most four shapes.
You may not use the letter E.
I say I like writing science fiction, but the truth is that I like some subject-matter constraints that mostly restrict me to the "science fiction" genre. But I don't mind, because "science fiction" sets a lot of presets for me that I don't have to change unless I want to.
No, not really. Actually making a painting, or a story, or a computer program, requires making decisions. Billions of decisions. You don't have time to make these decisions individually. But the right constraint can really cut it down, and let you hunt through a much smaller set for some good ideas.
In fact, to write is to continually add constraints. Every word you write reduces the options you have for the next word. Most of the time this makes things easier. Sometimes a word cuts off too many of you options. You killed a character you don't want to kill. You have to remove a constraint and backtrack.
This is the second thing I don't want the aliens to hear until they're older, because it's like telling little children how they were made. In fact, it's kind of the same story.
Stop looking for the absolute best constraint. Pick one at random and see if it works.
This may seem like a Dadaist fantasy, but it's a time-honored strategy.
In a deterministic universe there is no randomness. But we can make it look like there is randomness, and we can use randomness to bootstrap a creative process. I think this is the closest thing to magic that we'll ever see in the real world.
This isn't helpful. Pure randomness is as boring and as terrifying as a blank page. So just saying "randomness" doesn't solve the problem either. We thought randomness would help us choose constraints, but now we need to put constraints on the randomness to find the constraints on our work.
We need a deck of cards. Oblique Strategies, One Upon A Time, and the Tarot all use cards to put constraints on pure randomness.
I'm going to tell you about three kinds of aliens I've seen in the wild, and created myself. I differentiate them based on how they combine constraints with randomness.
That's not the whole comic, but you get the idea. Pokey the Penguin is usually classified as a 'surrealist' or 'dadaist' comic because it gets laughs through unpredictability and absurd juxtapositions, rather than punchlines or character humor.
However, it's not really that unpredictable. A comic strip is made of panels, and when you read it, your brain creates implicit transitions between one panel and the next. So it's fine. In panel five you don't wonder where the film projector came from. You know there was a scene change. Pokey the Penguin has a structure by virtue of being a comic strip. It's the same structure as my slide deck: a linear sequence of images.
Pokey the Penguin is weird, but it's not an alien, it's a work of art created directly by a human being. However, back in 1998 I wrote my first creative alien, based on Pokey the Penguin and called 'Dada Pokey'.
You can still read this. Your brain still fills in the blanks between the panels. The blanks are a lot bigger, but it's still pretty legible. It's some kind of story where two penguins have a jetpack accident and Superman saves them. Except I didn't write this. An alien wrote it using the constraints I set up. And the way in which it makes sense is a weird, alien kind of way that's interesting to me as a science fiction fan.
1. Because we are hard-wired to construct a narrative out of whatever little bits of information we see.
2. This particular comic is short. Any three points form a triangle. If this comic were longer than three panels it would fall apart pretty quickly.
and 3. By virtue of being a comic strip, Pokey the Penguin already contains a lot of implicit structure, which Dada Pokey preserves. I don't cut up the panels and recombine them into new images; I just shuffle the panels.
A Duchamp alien is the software equivalent of a deck of cards. It takes a deck of cards, shuffles it, and deals some out. There's no structure other than the fact that everything's on a card and the cards are the same size.
Markov chains are used in a lot of scientific applications to model sequential processes. If you think of a text as a sequence of words, you can use a Markov chain to predict the next word of a text based on the previous word.
So if you tell a properly trained Markov chain that you just saw "I", it will make a guess, and most of the time it will guess something simple like "I am" or "I had". Occasionally it will guess something more unusual like "I love" or "I realize". And very rarely it will guess something strange like "I robot" or "I Claudius".
There's a dial on a Markov chain which you can tweak. The dial is called 'order' and it basically controls how much of the original text the Markov chain tries to match when it guesses. This here is an order one Markov chain, meaning it looks at only one word: the previous word.
There's a little bit of Lovecraft, then the alien sees some overlap between Lovecraft and King James and it switches to King James for a while, then it switches back. It's pretty good.
But the idea of using Markov chains for artistic purposes goes back at least to 1972, and although the first Markov chain was used in a mathematical proof, it was trained on a work of literature.
Anyway, 1913, and the tsarist government put out a call for celebrations of the 300th anniversary of the House of Romanov.
Markov was an anti-tsarist, and as a big "screw you" to the House of Romanov, he set up a seminar celebrating a milestone he thought was more important: the 200th anniversary of Jacob Bernoulli's founding work on probability theory.
See, Markov's colleague Nekrasov had published a paper that contained a mathematical argument for free will. Markov saw this paper and decided that it was clobberin' time. He noticed that Nekrasov's argument relied on an unproven assumption about Jacob Bernoulli's Law of Large Numbers.
All he had to disprove this assumption was invent Markov chains. And then he had to create the world's first actual Markov chain, which meant spending several solid days counting the vowels and consonants in Pushkin's poem Yevgeny Onegin.
I bring this up for two reasons. First, it's one of the most interesting stories I've heard recently. I took this from a talk Brian Hayes gave; he also wrote about it for American Scientist.
When you run them twenty thousand times, Markov chains become very predictable. This is actually why Markov invented them--this is the fact that disproves Nekrasov's unproven assumption about the Law of Large Numbers. But it's not so great if you intend a Markov alien to be creatively satisfying.
A Markov chain with a high order generates interesting results if you give it several different source texts, as Charles Stross did. It will jump from one text to another in a sort of stream of consciousness mashup. And that's pretty cool, it's a lot cooler than endless duckspeak, but it does eventually gets old.
This is the problem I have with Markov chains. They're really good for creating two specific effects, but nerds tend to treat them like the go-to technique when you want to make an alien.
Here's the first time I invented it, in 2009. I gave the Dada Pokey treatment to another of my favorite web comics, Dinosaur Comics. Here's a Dinosaur Comics strip.
Then I deal the appropriate number of cards from each deck. I pick a first panel from one comic, a second panel from another comic, a third panel from yet another comic, and so on.
I wrote a script that slices up all the Dinosaur Comics images on their panel borders, chooses a new set of panels, and stitches them back together into a new image. Ta-da!
Anyhow, once I was looking for a technique like this, I quickly rediscovered it. It didn't seem to have a name, so I call it "Queneau assembly", after the French writer Raymond Queneau, who as far as I know was the first person to apply it to text.
So instead of turning the page and reading a completely different sonnet, you can peel back the page for just the first line of the poem, and get a sonnet that's exactly the same except for the first line.
I think these are the big three, creatively, right now. Duchamp, Markov, Queneau. There's another technique, using generative grammars, which isn't used much anymore, and I'm sure there are more types of aliens waiting to be discovered.
Right now Queneau assembly is my favorite technique. You could argue that it's the same as the Duchamp technique, because it's just dealing out cards from different decks. But with Queneau assembly so much of your time is devoted to describing the structure of the thing you're varying randomly that I call it a separate technique.
This is not because I have any particular love for Twitter. I'd much rather keep everything on my own website, which I control. But remember that second front of the artificial intelligence project, in which humans are trained to act in ways that are legible to aliens? This project has leveled the playing field somewhat between humans and aliens, and Twitter is the best place to exploit this.
1. Every message on a social media site has an author, and every author is implicitly a person. It makes no sense to talk to the Pepsi website, but you can talk to "Pepsi" on Twitter. Every nonhuman on Twitter—a brand, or an alien, or whatever else—gets anthropomorphized and treated the same as a human.
2. Twitter has lots of users. The users interact with each other, and nonhuman users are subject to more or less the same social expectations as human users.
And here's the big one. this is also where Twitter differentiates itself from Facebook and other social media sites.
3. Twitter's onerous 140-character constraint puts human-level performance within reach of an alien.
#3 is the most important one. The longer you look at the output of a Markov chain, the less sense it makes. The longer a Dada Pokey gets, the more conceptual leaps you have to make to fill in the action between panels, and at some point you'll just say "fuck it" and quit.
But you can get around this problem by drastically shortening the length of the interaction. You don't have long conversations with your phone, and 140 characters isn't a novel. It's not that hard to create a software alien that can imitate a human being for 140 characters. There are hundreds of interesting ways to do this. The restriction that, from one point of view, was so extreme that I pitched you a satirical dystopia about it, becomes an artistic boon.
If you followed Tetsuo back when he was posting on Twitter, his posts would show up in between your friends complaining about whatever people complained about in 2011. And people retweeted him, and some people talked back to him, the way you talk back to the television. This was a real revelation for me, because it brought the aspect of a performance to my writing.
Note all the little things I added to create the appearance of a personality for this alien. It's not just a list of product ideas, one after another. It's a person, a sketch comedy character, "The Serial Entrepreneur", this tinkerer in a garage who writes ideas down on napkins and is really excited about ideas that never work.
"What type of carbon is a dice? A die-amond."
Even though Joking Computer is explicitly a computer program, and the web site about the underlying research treats it as a computer program, because it has a Twitter account it's presented as this goofy anthropomorphic computer in foolscap.
Serial Entrepreneur and Joking Computer come up with jokes that are funny because they're bad in a creative way. AmIRite Bot is different; it says "my joke may be bad, but it's better than the joke you came up with, puny human."
The space between these two initially unrelated things becomes the space between panels in a comic strip. You fill it in mentally.
And again, this works because "social media" is a relatively level playing field. Everything you do on social media is split up into little chunks that go into a database. Computers are great at putting things into databases, and humans are great at filling in the blanks when it looks like another human is talking to them.
But @Horse_ebooks slowly acquired a large following, because it turns out that once in a while, bibliomancy strikes gold. Picking random quotes and presenting them out of context occasionally generates something really funny.
Like the Markov chain, @Horse_ebooks was created by a Russian, Alexei Kouznetsov, for less than honorable reasons. But he'd created an alien that—thanks to the human tendency to fill in the blanks—appeared to have occasional flashes of brilliance. And so the reputation of this spambot grew and grew. It was as though Marvin the Martian had come to conquer Earth, but the language in which he delivered his threats was so beautiful that he won our hearts as a great poet.
And then, on September 14, 2011, about a month before my Constellation Games characters started posting to Twitter, humanity struck back. @Horse_ebooks was... posessed.
That first slide of quotes I showed you, that was the 'best of'. Those quotes were separated by days. This slide shows four consecutive posts from @Horse_ebooks in 2013. They're all good.
This was remarked upon at the time, and in retrospect, it was obvious what had happened. Someone was now manually curating the output of the horse_ebooks algorithm, picking out the good stuff. @Horse_ebooks had become a pod person. Everyone thought it was a goofy blustering Marvin the Martian alien, but it was a human in a mask. Humanity had invaded the aliens.
After about two years, the humans took off the mask. The guys who'd bought @Horse_ebooks announced that it was all part of an art project that tied into their Alternate Reality Game. They were tired of impersonating the alien, tired of getting up in the middle of the night to type in quotes.
Reactions to this were complicated but generally negative. Some people were disappointed. Some were cynical, some were angry.
I'd spent over ten years making these little aliens, and over a year giving them personalities and hooking them up to Twitter. And @Horse_ebooks—the @Horse_ebooks people thought they were getting—it's not all that complicated. It's easy to pull short random quotes out of books. Some of these quotes turn out to be funny, some of them turn out not to be funny. How difficult is it to find the linguistic features that make this sort of thing funny?
It can't be that hard. From an AI standpoint, the Joking Computer is actually more impressive. So... could I give people the @Horse_ebooks they wanted? Could I turn the fake alien that people had been hoping for into a real alien?
It doesn't have 200,000 followers, and it never will, but it is a faithful recreation of something that turned out to be a hoax: an alien that has a sense of humor despite not understanding English.
You might say that this year's Foolscap is the first science fiction convention to have guests who are aliens. But then you'd be anthropomorphizing a piece of software. And anyway, who knows? If you were an alien, there are a lot worse places you could hang out.
Constellation Games, quoted and used as an example text several times, is my first novel. It talks a lot about alien and machine intelligence.
I slightly edited and modified this talk for the Web, including adding a lot of hyperlinks.
In computer science, artificial intelligence is the stereotypical failed project.
We have been working on this for sixty years and we don't have anything that approaches the general-purpose cognitive ability of a human being.
As a science fiction writer, I think it's presumptuous to use a human being as the standard of intelligence. We don't expect space aliens to have the general-purpose cognitive ability of a human being. We expect them to be different from us.
If you look at artificial intelligence as the project of creating intelligent, strangely obsessive aliens, the situation looks a lot better. There's software that can beat you at chess, or at Jeopardy. You can ask a computer to give you directions from one place to another, and the computer will figure it out and send you a map. What goes on inside the software has almost nothing to do with how a human brain solves the same problem, but the software is better at its task than any human.
It's easy to make software that claims to have human attributes, so we have a tendency to anthropomorphize software, even software that's very stupid. The way we dress our dogs in little costumes. If you take this to its logical conclusion, it's hard to escape the conclusion that we have created a race of servants, if not outright slaves. Our software has intelligence, but no agency.If you habitually talk to your phone, think about the sort of things you say to it. Would you talk to another human like that? What would happen if you asked your phone about itself? You can try it. You'll probably get a canned, jokey response that says nothing. The sort of response a well-trained butler would give.
So I would prefer not to anthropomorphize pieces of software. But I can't stop, because I'm a fiction writer. I anthropomorphize everything. I have to anthropomorphize a fleeting idea in my head, and then I have to write down, like, ten or fifteen words of description, and I have to convince you that behind those words there is an independent intelligence with a rich inner life.
Besides which, I'm a science fiction writer, and if I can anthropomorphize pieces of software, I've got some characters.
So I've been trying out the idea of thinking of pieces of software as alien beings of various types. It's a fun character, almost a stock character. They're different from us. They understand English to varying degrees. None of them really understand humanity, but that doesn't necessarily mean they're stupid. Maybe we have something to learn from each other.
The artificial intelligence project has split into two smaller projects. The first project is to create servants: software that can interact with humans, on the humans' terms. This has been a big success. The software that creates maps for you works very well, and it's operating on your terms. The computer doesn't need maps. It's stuck in a data center in Idaho.The second front in the AI project is not an official research project, but it has also been a big success. This is the project to make humans more computer-legible. To train us to act in ways that can be processed by these weird, obsessive aliens that don't really understand humanity.
When we call the bank, an alien answers the phone. We communicate with the alien by pushing buttons on our phone, or by shouting words one at a time, as if we were trying to get a French person to understand French. And if it doesn't work, we demand to talk to a human. But when we do, we talk to a human who's been trained with a script to act only in certain, pre-programmed ways. The human is being paid to pretend to be an alien who speaks English really well.
When I write fiction, I frequently write the first draft in longhand. This is bad for aliens, because they can't read my handwriting. And that's bad for me, because the entire publishing process is a partnership between humans and aliens. To get my writing to people who want to read it, at some point in the process I have to go back and type up my prose and put it into a structured form that an alien can consume.That's not so bad, because the aliens don't read. They don't care about the content of my fiction. I can write whatever I want, so long as I think I can convince a human to publish it and a bunch of other humans to read it.
But here's the weird thing: sometimes I write especially for an alien called Twitter. And when I do this, the alien's biology has a huge effect on what I write.You can think of Twitter as an alien library that can contain an infinite number of texts. Due to a quirk of its biology, it cannot metabolize any text longer than 140 characters.
When I write something for Twitter I identify myself and I type into a text box. Whatever I typed is stored in the library. A lot of incidental information is stored alongside it: the current time, the fact that it was me who wrote this, and my physical coordinates.
This is Twitter's food. It can only accept information that fits this specific format. If I want to feed it something that's 141 characters long, I can't do it. That last character has to go into a separate piece of text, encrusted with its own set of metadata. This is a biological limitation. Twitter cannot contain a 141-character text any more than I can fly.
This biological limitation changes the way I write. I use abbreviations I would normally never use. I reword things to make them shorter, even when that makes them sound stilted. Sometimes I can't get it to work, so I give up and don't write anything after all.
We ended up splitting artificial intelligence into two projects because the aliens we've created are very good at handling some kinds of data and very bad at handling others. It's the same for humans. We're good at language processing and general-purpose reasoning, and we're bad at everything else.What we see now is a kind of symbiosis, where the aliens make maps and remember things for us, and in return we only do things that the aliens can understand. And of course we fix the aliens when they break.
Unfortuantely, because the aliens are so obviously stupid by human standards of intelligence, we continually underestimate them. Twitter doesn't understand English; it only counts characters. It doesn't understand the things we're feeding it. But it has a very good idea of who is feeding it, when those people are awake, where they live and work, and the relationships they have with each other.Okay, I love aliens. I create aliens all the time. Real aliens made of software, like Twitter, and fictional aliens made of words, like the ones in my novels. And fictional pieces of software created by fictional aliens. But this thing that knows a lot about our relationships yet doesn't understand English, doesn't sound like a trustworthy alien. At the very least, it sounds like an alien that can be easily tricked into doing a lot of damage. And we've discovered that frequently, in the shadows, there lurk sinister humans—the secret masters of the aliens.
We feed these aliens without even knowing. For instance, I'm giving a talk right now, and this talk has slides. I've done a lot of work ahead of time to organize my talk into a sequence of discrete elements, each containing an outline. The title of the slide contains the main point of the slide. The top-level bullet points explain that idea, and when necessary I put subordinate points below the top-level bullet points.In computer science we call this a data structure: specifically, it's a list of trees. You are experiencing my talk as freeform English prose, but I've also unwittingly structured it for consumption by a machine. A machine that probably doesn't understand the sentence "This talk has slides", but which now knows that "This talk has slides", whatever that is, is a more important concept to my talk than "(A list of trees)".
For this reason I'm more or less going to stop showing slides like that in my talk. I'm going to be saying some things that I don't want software-based aliens to hear until they're a little older.
Now, all of this is a gold mine for science fiction writers. We love aliens and computers, but we also love to pit a individual against a system. A bureaucracy, a government, a corporation, a religion, a family. A system that imposes constraints on people, and a person who finds those constraints dehumanizing to the point of being intolerable.
So I'm going to pitch to you a story idea.That's pretty silly. There's no actual conflict and it's kind of a clumsy satire. But it's pretty similar to, say, Harrison Bergeron, which is a classic. It's a timeless story, really. Some external authority has imposed a ridiculous restriction on everybody, and one character has the courage to buck the system. Yay individualism!
Okay, but.
Here's something I don't want the aliens to hear until they're older: we, their human parents, are not so smart, either. All that natural-language text they don't understand? Most of it is not very interesting. We tell the same jokes over and over. We mistake cliches for wisdom. We say things just to hurt people we don't know. We talk just to hear our own voices.This is why we build aliens like Twitter. There is something appealing about constraints. We thought "maybe if we only had 140 characters, we'd be forced to make every one count." On the same logic that if you knew you only had a month to life, you'd get off the couch.
In real live, the 140-character constraint didn't solve any of the problems I just mentioned. If you want to say things that will hurt people you don't know, Twitter is probably the best medium ever invented. But this oppressive, fascist-seeming constraint has led to a breakthrough in human creativity and in human-alien relations.
How is this possible? We're American science fiction fans! We hate constraints! And it's true that constraints that are forced on you are terrible. But voluntarily-chosen constraints are the source of creativity.
Fiction writers should know this better than anyone. When we stare at the blank page or the empty word processor window, that's the moment when we are completely free, subject to no laws, not even the laws of nature. We are Harrison Bergeron, soaring high above all the normals. But we hate that moment! It's paralyzing. We need a plotline, or a character, or a genre, or at least an opening sentence, or else we'll stare at that blank page all day.
Earlier I mentioned that humans are good at language processing and general-purpose reasoning, and bad at everything else. I originally also had "creativity" on that list, but I don't know whether creativity is really a skill, and I also don't think we're all that good at creativity. Creativity comes from picking the right constraints.
Here's a constraint for a painter:You may only paint straight lines. You may only use the primary colors: red, yellow, and blue, plus black, white, and grey.
These are Piet Mondrian's constraints, and he was so successful at it that these constraints kind of became the national constraints of the Netherlands.
Here's another severe constraint for a painter:You may only paint rectangular shapes in solid colors. A single canvas may contain at most four shapes.
This is Mark Rothko's constraint. His paintings make people cry.
Here's a constraint for a writer:You may not use the letter E.
In 1969 Georges Perec wrote a novel called La Disparition without using the letter E. This sort of constraint is so common it has its own name: a lipogram. Unlike with Mondrian and Rothko, I don't think La Disparition is great art. But all writers pick some kind of constraints, or they can't do their work.I say I like writing science fiction, but the truth is that I like some subject-matter constraints that mostly restrict me to the "science fiction" genre. But I don't mind, because "science fiction" sets a lot of presets for me that I don't have to change unless I want to.
Why would you subject yourself to these dorky constraints? On top of the ordinary constraints like 'put paint on a canvas' or 'write complete sentences that are mostly grammatical.' Hell, why not get rid of those? Isn't art all about breaking boundaries?No, not really. Actually making a painting, or a story, or a computer program, requires making decisions. Billions of decisions. You don't have time to make these decisions individually. But the right constraint can really cut it down, and let you hunt through a much smaller set for some good ideas.
In fact, to write is to continually add constraints. Every word you write reduces the options you have for the next word. Most of the time this makes things easier. Sometimes a word cuts off too many of you options. You killed a character you don't want to kill. You have to remove a constraint and backtrack.
But, this little writing-group pep talk doesn't actually solve anything, because you still have to make these decisions. You have to pick the right constraints. How are you supposed to do that, smart guy?This is the second thing I don't want the aliens to hear until they're older, because it's like telling little children how they were made. In fact, it's kind of the same story.
The answer is randomness.Stop looking for the absolute best constraint. Pick one at random and see if it works.
This may seem like a Dadaist fantasy, but it's a time-honored strategy.
Here's Brian Eno and Peter Schmidt's 'oblique strategies' deck. There are a bunch of constraints on a deck of cards. If you're stuck, creatively, you draw a card, and it gives you a different way of looking at the problem.
Here's a popular game that uses randomness to get people to come up with stories. Fully formed, creative stories from people who aren't professional writers. These stories aren't going win the World Fantasy Award, but there's no agonizing in front of a blank page. You say whatever you need to say to get rid of your cards, and you have a good time.
This is part of a tarot deck painted by my friend Beth Lerman. Tarot is a technique for using randomness to make sense of your everyday life. This technique is hundreds of years old and millions of people have used it.In a deterministic universe there is no randomness. But we can make it look like there is randomness, and we can use randomness to bootstrap a creative process. I think this is the closest thing to magic that we'll ever see in the real world.
But even pure randomness is not enough. Here's pure randomness.This isn't helpful. Pure randomness is as boring and as terrifying as a blank page. So just saying "randomness" doesn't solve the problem either. We thought randomness would help us choose constraints, but now we need to put constraints on the randomness to find the constraints on our work.
We need a deck of cards. Oblique Strategies, One Upon A Time, and the Tarot all use cards to put constraints on pure randomness.
Now we come back to aliens. I've been making a lot of software projects recently, unfortunately at the expense of my prose writing. Little aliens whose purpose is not to gather data or carry out tasks for humans, but to produce works of art. It's an interesting project, because 100% of the creativity involved is in finding the right constraints to put on randomness. Once you figure that out, you can rely on randomness to produce the art.I'm going to tell you about three kinds of aliens I've seen in the wild, and created myself. I differentiate them based on how they combine constraints with randomness.
The first kind of alien I name after Marcel Duchamp, the patron saint of Dadaism. Duchamp was a much more complicated character than "patron saint of Dadaism" would imply, but I think he deserves to have this named after him.
So, there's a web comic I like called Pokey the Penguin. It's been around since 1998. Here's one of the recent comics.That's not the whole comic, but you get the idea. Pokey the Penguin is usually classified as a 'surrealist' or 'dadaist' comic because it gets laughs through unpredictability and absurd juxtapositions, rather than punchlines or character humor.
However, it's not really that unpredictable. A comic strip is made of panels, and when you read it, your brain creates implicit transitions between one panel and the next. So it's fine. In panel five you don't wonder where the film projector came from. You know there was a scene change. Pokey the Penguin has a structure by virtue of being a comic strip. It's the same structure as my slide deck: a linear sequence of images.
Pokey the Penguin is weird, but it's not an alien, it's a work of art created directly by a human being. However, back in 1998 I wrote my first creative alien, based on Pokey the Penguin and called 'Dada Pokey'.
Here's an example. Dada Pokey takes the image files from Pokey the Penguin, shuffles the deck, and deals out a new hand, in the same linear format as the original comic.You can still read this. Your brain still fills in the blanks between the panels. The blanks are a lot bigger, but it's still pretty legible. It's some kind of story where two penguins have a jetpack accident and Superman saves them. Except I didn't write this. An alien wrote it using the constraints I set up. And the way in which it makes sense is a weird, alien kind of way that's interesting to me as a science fiction fan.
Now, if the alien chose these panels completely at random, which I assure you is the truth, why is this story readable? Why doesn't it look like white noise?1. Because we are hard-wired to construct a narrative out of whatever little bits of information we see.
2. This particular comic is short. Any three points form a triangle. If this comic were longer than three panels it would fall apart pretty quickly.
and 3. By virtue of being a comic strip, Pokey the Penguin already contains a lot of implicit structure, which Dada Pokey preserves. I don't cut up the panels and recombine them into new images; I just shuffle the panels.
A Duchamp alien is the software equivalent of a deck of cards. It takes a deck of cards, shuffles it, and deals some out. There's no structure other than the fact that everything's on a card and the cards are the same size.
The second kind of alien is the one you will probably get if you ask a random nerd about this problem. I call them Markov aliens because they're based on a mathematical technique called the Markov chain.Markov chains are used in a lot of scientific applications to model sequential processes. If you think of a text as a sequence of words, you can use a Markov chain to predict the next word of a text based on the previous word.
We do this by looking at word frequencies in some original source text. For instance, in English, the word "I" is frequently followed by the word "am". Or the word "had". Or one of many more specific verbs like "love" and "realize". It's very rare for the word "I" to be followed by a noun, like "robot" or "Claudius", but it does happen sometimes.So if you tell a properly trained Markov chain that you just saw "I", it will make a guess, and most of the time it will guess something simple like "I am" or "I had". Occasionally it will guess something more unusual like "I love" or "I realize". And very rarely it will guess something strange like "I robot" or "I Claudius".
A Markov chain becomes an alien when you feed its output back into its input. Now you can keep it going for more than one word. You say "I" and the Markov chain guesses "had". You say "had" and the Markov chain guesses "never". You say "never" and the Markov chain guesses "mind". The output: "I had never mind." An alien thing to say.
A Markov chain trained on text will generate what can be charitably referred to as plausible seeming gibberish. Here's an example of a Markov chain trained on the text of my novel Constellation Games.There's a dial on a Markov chain which you can tweak. The dial is called 'order' and it basically controls how much of the original text the Markov chain tries to match when it guesses. This here is an order one Markov chain, meaning it looks at only one word: the previous word.
Here's an order two Markov chain. At this point more of the original text starts showing up, and the sentences become a little more coherent. But you, the reader, are still doing a lot of the work. For instance, the two references to beer in this paragraph are coincidental. By the time it gets to the word "fires", this Markov chain has completely forgotten that it ever mentioned the word "beer".
If you turn up the dial too high, your Markov chain will just quote the original text obsessively, like a Monty Python fan. Here's the output of an order three Markov chain. It's pretty much a direct quote from the novel.
You may have seen recently a Markov alien by Charles Stross, building on work by Michael Walker. He wrote a Perl script that mashes up the King James Bible and the works of H.P. Lovecraft. Here's some output.There's a little bit of Lovecraft, then the alien sees some overlap between Lovecraft and King James and it switches to King James for a while, then it switches back. It's pretty good.
But the idea of using Markov chains for artistic purposes goes back at least to 1972, and although the first Markov chain was used in a mathematical proof, it was trained on a work of literature.
Markov chains were invented in pre-revolutionary Russia by A. A. Markov, a mathematician who liked picking fights. The two fights we're interested in right now happened in 1913, when he was retired. So he was less in this phase of his life and
closer to this phase.Anyway, 1913, and the tsarist government put out a call for celebrations of the 300th anniversary of the House of Romanov.
Markov was an anti-tsarist, and as a big "screw you" to the House of Romanov, he set up a seminar celebrating a milestone he thought was more important: the 200th anniversary of Jacob Bernoulli's founding work on probability theory.
At this seminar, he announced the discovery of what would prove to be one of the most useful mathematical tools of the twentieth century. Why did he come up with this brilliant idea? Apparently it was to prosecute a feud against a fellow mathematician!See, Markov's colleague Nekrasov had published a paper that contained a mathematical argument for free will. Markov saw this paper and decided that it was clobberin' time. He noticed that Nekrasov's argument relied on an unproven assumption about Jacob Bernoulli's Law of Large Numbers.
All he had to disprove this assumption was invent Markov chains. And then he had to create the world's first actual Markov chain, which meant spending several solid days counting the vowels and consonants in Pushkin's poem Yevgeny Onegin.
I bring this up for two reasons. First, it's one of the most interesting stories I've heard recently. I took this from a talk Brian Hayes gave; he also wrote about it for American Scientist.
The second reason I bring it up has to do with one of my big literary influences, the Illuminatus! trilogy. In that story there's a character named Markoff Chaney who goes around sowing chaos by putting up official-looking but misleading signs. The obvious connection between the character's name and the random component of Markov chains is made explicit in the book, but I also think there's a kinship between Markoff Chaney's bitter, lifelong campaign against authority and the real-life feuds of Andrey Andreyovich Markov.
If a Duchamp alien is shuffling cards and dealing out hands, then a Markov alien is playing some kind of Exquisite Corpse-type game. When it puts down a card it's always trying to match the previous card. Markov chains have a probabilistic component, but the probabilities are weighted. Otherwise "I robot" would be just as likely as "I had".When you run them twenty thousand times, Markov chains become very predictable. This is actually why Markov invented them--this is the fact that disproves Nekrasov's unproven assumption about the Law of Large Numbers. But it's not so great if you intend a Markov alien to be creatively satisfying.
A Markov chain with a low order reliably generates what George Orwell called duckspeak: it looks like it should make sense, but it's gibberish. Generative duckspeak is great for mocking the platitudes and euphemisms spewed forth by The Management, as Markoff Chaney did with his misleading signs. But it's very unlikely anything interesting will happen in your brain if you read it.A Markov chain with a high order generates interesting results if you give it several different source texts, as Charles Stross did. It will jump from one text to another in a sort of stream of consciousness mashup. And that's pretty cool, it's a lot cooler than endless duckspeak, but it does eventually gets old.
This is the problem I have with Markov chains. They're really good for creating two specific effects, but nerds tend to treat them like the go-to technique when you want to make an alien.
A couple years ago I announced that I was sick of Markov chains, and I set out to find another way to make an alien. I discovered that I'd already invented one. In fact, I invented it three times, and other people had invented it before me.Here's the first time I invented it, in 2009. I gave the Dada Pokey treatment to another of my favorite web comics, Dinosaur Comics. Here's a Dinosaur Comics strip.
Here's another one.
Every Dinosaur Comics has exactly the same graphics. The second panel always has that close-up of T-rex, the third panel always has him about to stomp on a house, and the last panel always has him going like this. [GOES LIKE THIS]
The Dada Pokey technique won't work here, because the panels are not interchangeable. The cards are all different sizes. But I can still shuffle Dinosaur Comics. I just need to create several decks of cards in a way that respects the underlying structure.Then I deal the appropriate number of cards from each deck. I pick a first panel from one comic, a second panel from another comic, a third panel from yet another comic, and so on.
I wrote a script that slices up all the Dinosaur Comics images on their panel borders, chooses a new set of panels, and stitches them back together into a new image. Ta-da!
This is similar to a traditional children's activity called "heads, bodies, legs". Here's a "heads, bodies, legs" book from 1980, by Helen Oxenbury. Note the number in the title. 729 is nine to the third power. There are nine heads, nine bodies, and nine legs. Since they're all on their own individual pages, you can combine them in 93 ways.Anyhow, once I was looking for a technique like this, I quickly rediscovered it. It didn't seem to have a name, so I call it "Queneau assembly", after the French writer Raymond Queneau, who as far as I know was the first person to apply it to text.
Here's Raymond Queneau. In 1960 he was a founder of the Oulipo circle, a French literary movement which also included George Perec of not-using-the-letter-E fame. They were all about creating stuff under self-imposed constraints, but they didn't have computers, so they wrote novels and poems instead of making creative aliens.
The Oulipo circle was created because Raymond Queneau needed help creating a heads-bodies-legs book for poetry. The idea is, you open the book and you read a ten-line sonnet on the first page. But the page is perforated into strips, like a flyer on a telephone pole. Every line of the sonnet is effectively on its own individual page. Head, body, legs.So instead of turning the page and reading a completely different sonnet, you can peel back the page for just the first line of the poem, and get a sonnet that's exactly the same except for the first line.
There are ten variants for each line, and they all fit the same meter and the same rhyme scheme, so there are 1010 possible sonnets in one little book. Queneau called it Cent mille milliard de poemes. I would have called it 1010 Poems, but okay.
Queneau assembly is great because it reproduces a preexisting structure without stretching the size of the weird gap you have to leap to get from one part of the structure to the next. You saw it work for Dinosaur Comics. Queneau proved that it works for sonnets. It also works for limericks: here are some examples from a Queneau alien I created.
Here's an example from an alien I created that invents new board games. Both the title and the description are generated with Queneau assembly.I think these are the big three, creatively, right now. Duchamp, Markov, Queneau. There's another technique, using generative grammars, which isn't used much anymore, and I'm sure there are more types of aliens waiting to be discovered.
Right now Queneau assembly is my favorite technique. You could argue that it's the same as the Duchamp technique, because it's just dealing out cards from different decks. But with Queneau assembly so much of your time is devoted to describing the structure of the thing you're varying randomly that I call it a separate technique.
Let me bring this back to Twitter. For fifteen years I've been making projects like Dada Pokey and the Dinosaur Comics thing and putting them up on my website. But as of recently I'm really into Twitter. In 2013 I put three projects like this up on my website, versus eight projects on Twitter.This is not because I have any particular love for Twitter. I'd much rather keep everything on my own website, which I control. But remember that second front of the artificial intelligence project, in which humans are trained to act in ways that are legible to aliens? This project has leveled the playing field somewhat between humans and aliens, and Twitter is the best place to exploit this.
1. Every message on a social media site has an author, and every author is implicitly a person. It makes no sense to talk to the Pepsi website, but you can talk to "Pepsi" on Twitter. Every nonhuman on Twitter—a brand, or an alien, or whatever else—gets anthropomorphized and treated the same as a human.
2. Twitter has lots of users. The users interact with each other, and nonhuman users are subject to more or less the same social expectations as human users.
And here's the big one. this is also where Twitter differentiates itself from Facebook and other social media sites.
3. Twitter's onerous 140-character constraint puts human-level performance within reach of an alien.
#3 is the most important one. The longer you look at the output of a Markov chain, the less sense it makes. The longer a Dada Pokey gets, the more conceptual leaps you have to make to fill in the action between panels, and at some point you'll just say "fuck it" and quit.
A computer can't write a novel. A novel is too long. We had a contest last year, NaNoGenMo, and although my entry was clearly the best, it was a) objectively terrible, and b) a big rip-off of some public domain novels.But you can get around this problem by drastically shortening the length of the interaction. You don't have long conversations with your phone, and 140 characters isn't a novel. It's not that hard to create a software alien that can imitate a human being for 140 characters. There are hundreds of interesting ways to do this. The restriction that, from one point of view, was so extreme that I pitched you a satirical dystopia about it, becomes an artistic boon.
I discovered this in 2011 when I actually put a space alien on Twitter. Constellation Games was being serialized, one chapter a week, so as a promotion/bonus I created two scripted Twitter accounts for characters in the novel, which ran alongside the serialization. One of the feeds was for the book's main character, Ariel Blum. His Twitter feed was mainly a chance for me to use a bunch of ideas that wouldn't fit in the book. It looks like normal Twitter posts from an average computer programmer who happens to be living after first contact.
But the other Twitter feed was for Tetsuo Milk, an alien anthropologist, and it's not like normal Twitter at all. Tetsuo is learning how to navigate human civilization. He's a complete extrovert who's always wrong about the people he's interacting with.
And Tetsuo has a way of speaking English that's kind of the opposite of a Markov chain. It looks like gibberish, because he keeps using the wrong word, but if you look at it you can figure out what he means. I think this is the perfect combination for Twitter, because you can't let Tetsuo talk for too long. It's too confusing. He has one big speech late in the book, but I could justify it by saying that by that time his English is a little better.If you followed Tetsuo back when he was posting on Twitter, his posts would show up in between your friends complaining about whatever people complained about in 2011. And people retweeted him, and some people talked back to him, the way you talk back to the television. This was a real revelation for me, because it brought the aspect of a performance to my writing.
Now it's 2014 and there are a ton of aliens on Twitter. I've got one more big story to tell you but I want to sort of take you on a tour of these friendly aliens and explain how they work.
In my opinion @MarkovChocolate is one of the only good Markov aliens on Twitter. It uses a Markov chain to come up with types of chocolate candies. It was created by Tom Armitage.
Serial Entrepreneur is similar. It comes up with product ideas. It uses a generative grammar, and it's one of my most recent creations.Note all the little things I added to create the appearance of a personality for this alien. It's not just a list of product ideas, one after another. It's a person, a sketch comedy character, "The Serial Entrepreneur", this tinkerer in a garage who writes ideas down on napkins and is really excited about ideas that never work.
Joking Computer is actually a legitimate artificial intelligence project out of the University of Aberdeen. It has a sophisticated knowledge of word formation and synonyms which it uses to come up with the worst fourth-grade jokes you can imagine. Here's another one:"What type of carbon is a dice? A die-amond."
Even though Joking Computer is explicitly a computer program, and the web site about the underlying research treats it as a computer program, because it has a Twitter account it's presented as this goofy anthropomorphic computer in foolscap.
@AmIRiteBot is another automated comedian, by Darius Kazemi. It's a Duchamp alien. It takes Twitter's trending topics and replaces one word with a word that rhymes and then says "amirite?". Humans do this a lot on Twitter.Serial Entrepreneur and Joking Computer come up with jokes that are funny because they're bad in a creative way. AmIRite Bot is different; it says "my joke may be bad, but it's better than the joke you came up with, puny human."
@Unicode_ebooks is the purest Duchamp alien ever created. It's just random Unicode characters. @Unicode_ebooks was created by my friend Allison Parrish, and I hate it. It's so pointless. It's white noise.
I had some helpful suggestions for improving the look of @Unicode_ebooks, which Allison rejected, so I started a Markov vs. Nekrasov-type feud and created my own alien, @SmoothUnicode. I'm not sure how to classify it; it has some Duchamp elements and some Queneau elements. I think it's one of the best visual artists on Twitter, except for the ones that use images. However, it has about half as many followers as @Unicode_ebooks.
The Mixologist, by Patrick Rodriguez, comes up with cocktail recipes using a sophisticated Queneau-like technique, and names them using a simpler Duchamp technique. Again, note how a Twitter account that does nothing but display recipes is anthropomorphized into "The Mixologist".
The best alien on Twitter is probably Ranjit Bhatnagar's @pentamatron. It doesn't generate any text of its own. It is pure constraint. It retweets other people's posts to form an endless sonnet in iambic pentameter.
There are a number of aliens that work this way, by remixing things other people say. Ivy Faye's @AndNowImagine has a continuous search running for people who say "Imagine [whatever]". It picks two of them, and it combines them into a single narrative.The space between these two initially unrelated things becomes the space between panels in a comic strip. You fill it in mentally.
There's also my friend Rob Dubbin's @OliviaTaters, which might be the most effective of any alien at impersonating a real human on Twitter. I don't know what the algorithm is, exactly, but it mashes up little bits of phatic speech to make more phatic speech.
There's a ton of these aliens on Twitter—we call them 'bots', because we're nerds. If you have an idea that won't fit on Twitter, but you still want it to adopt a personality, you can use any other social media site. There's a YouTube bot called AutoVids that posts generated music videos. There are Tumblr bots that automatically create animated GIFs.And again, this works because "social media" is a relatively level playing field. Everything you do on social media is split up into little chunks that go into a database. Computers are great at putting things into databases, and humans are great at filling in the blanks when it looks like another human is talking to them.
Okay, here's my last story. This is a story about an alien invasion in reverse. It's a story about the complex relationship between humans and aliens, between natural intelligence and artificial intelligence. And it explains the two monitors I set up in the commons area. I call it...
"Invasion of the bot-y snatchers", or "Behold, a pale @Horse_ebooks".
So, @Horse_ebooks was a spambot on Twitter. Its job was to sell crappy ebooks about horses. It was very bad at this job. Here's its strategy: it posted quotes from the ebooks it was selling. Not good quotes, not the quotes you would see in a review. Just... random quotes from the middle of sentences with no context at all. It was a Duchamp alien all the way.But @Horse_ebooks slowly acquired a large following, because it turns out that once in a while, bibliomancy strikes gold. Picking random quotes and presenting them out of context occasionally generates something really funny.
Here are some of the best @Horse_ebooks quotes from the early era.Like the Markov chain, @Horse_ebooks was created by a Russian, Alexei Kouznetsov, for less than honorable reasons. But he'd created an alien that—thanks to the human tendency to fill in the blanks—appeared to have occasional flashes of brilliance. And so the reputation of this spambot grew and grew. It was as though Marvin the Martian had come to conquer Earth, but the language in which he delivered his threats was so beautiful that he won our hearts as a great poet.
And then, on September 14, 2011, about a month before my Constellation Games characters started posting to Twitter, humanity struck back. @Horse_ebooks was... posessed.
Two guys from New York, Jacob Bakkila and Thomas Bender, bought the Twitter account from the spammer. They shut off the automated posting and started posting manually. @Horse_ebooks was still posting the same stuff—truncated quotes from books—but the average comedic quality suddenly shot up. You didn't have to wait so long for something good. It was as if the dice suddenly started coming up boxcars every time.That first slide of quotes I showed you, that was the 'best of'. Those quotes were separated by days. This slide shows four consecutive posts from @Horse_ebooks in 2013. They're all good.
This was remarked upon at the time, and in retrospect, it was obvious what had happened. Someone was now manually curating the output of the horse_ebooks algorithm, picking out the good stuff. @Horse_ebooks had become a pod person. Everyone thought it was a goofy blustering Marvin the Martian alien, but it was a human in a mask. Humanity had invaded the aliens.
After about two years, the humans took off the mask. The guys who'd bought @Horse_ebooks announced that it was all part of an art project that tied into their Alternate Reality Game. They were tired of impersonating the alien, tired of getting up in the middle of the night to type in quotes.
Reactions to this were complicated but generally negative. Some people were disappointed. Some were cynical, some were angry.
We thought we'd found an alien. We thought a cliché science fiction story had come to life, that a computer on the Internet had become a strange form of intelligent life that could communicate with us. But no, that doesn't happen in real life. It was just a human feeding us a line in order to sell something, like always.
Now, people do a lot of things in the name of art, but why do this? Why would you laboriously pretend to be a robot for two years? I don't understand it. But the fact that I don't understand it told me something.I'd spent over ten years making these little aliens, and over a year giving them personalities and hooking them up to Twitter. And @Horse_ebooks—the @Horse_ebooks people thought they were getting—it's not all that complicated. It's easy to pull short random quotes out of books. Some of these quotes turn out to be funny, some of them turn out not to be funny. How difficult is it to find the linguistic features that make this sort of thing funny?
It can't be that hard. From an AI standpoint, the Joking Computer is actually more impressive. So... could I give people the @Horse_ebooks they wanted? Could I turn the fake alien that people had been hoping for into a real alien?
The details get kind of technical, but yes, I could. With a little help from Allison Parrish I reverse-engineered the @Horse_ebooks algorithm, including the part where a human picks out the good stuff, and I've created a Twitter account called @pony_strategies, which is basically the @Horse_ebooks spambot from the Constellation Games universe.It doesn't have 200,000 followers, and it never will, but it is a faithful recreation of something that turned out to be a hoax: an alien that has a sense of humor despite not understanding English.
That's where the big display monitor in the commons area comes from. It uses the same code as @pony_strategies, but it's taking quotes from online science fiction I got from Free Speculative Fiction Online. It represents about six thousand short stories and novels, from seventeen hundred authors.
The other monitor displays hapax legomena from the same corpus—words that only appear once across six thousand stories and novels. This is my little tribute to the moments of creation in the writer's life, when you come up with some turn of phrase or neologism that you know probably no one will appreciate on a conscious level.You might say that this year's Foolscap is the first science fiction convention to have guests who are aliens. But then you'd be anthropomorphizing a piece of software. And anyway, who knows? If you were an alien, there are a lot worse places you could hang out.
