Sheldon County is a podcast that will never sound the same twice. Every time someone listens to you, they start typing a random number on a website. This "seed" will launch a Rube Goldberg calculating machine that will create characters, relationships, jealousies, betrayals and maybe even a murder or two. These plot points will be converted into a text narration, read aloud with a speech synthesizer and then compressed into an audio file. Each time will be a unique version of the history of Sheldon County. A podcast made especially for you.
That's the dream anyway, the current execution still needs work. So far, there are only a few episodes of this podcast generated procedurally (you can listen to two below), and its creator, PhD student James Ryan, is still working on a website. He says that the backend software is mostly finished, but some final touches are needed, like creating a program to automatically add the music to each episode. "Right now I'm demonstrating the concept," he tells The Verge. "And then I have a dissertation to begin with."
In other words, it could be a time.
But what has been created so far is impressive and seems a small sample of the future. One where the content of entertainment is not only omnipresent (God knows that the world has enough podcasts) but it is also unique. As a way of doing fun things for humans, the content generated by procedures is not new, but it has become more complex in recent years. See, for example, videogames with generative elements such as No Man's Sky, which created unique planets for each player to explore; and Middle Earth: Shadow of War, which created enemies with elaborate stories that fascinated both the players and the story of the game.
Sheldon County seems basic in comparison, but that's only because its output is audio. The real mechanisms that create the characters and their interactions are much more complex. "It's a great research that goes beyond the limits of technology," says Mark Riedl, an associate professor in the Georgia Tech Entertainment Intelligence Lab, The Verge.
For Ryan, Sheldon County is the last step in a lifelong search to build computers that generate fictitious worlds. He is a linguist turned into a programmer, whose work with the Expressive Intelligence Studio of the University of California at Santa Cruz is dedicated to finding new ways of using machines expressively.
"When I learned to program, one of the first programs I did was a name generator," he says. "I would select two names from a list of thousands and combine them, and for me, creating this name was like creating a complete character, a small abstract person."
From this simple beginning, Ryan made increasingly complex world generators. Sheldon County itself is based on a program called Hennepin, which creates characters, their social networks and the world in which they live. Ryan compares Hennepin to "the world's largest Excel spreadsheet," with endless rows of cells that correspond to characters, traits, relationships, professions, etc.
& # 39; Sheldon County & # 39; is created by "the world's largest spreadsheet"
There is no visualization or textual output, only data. But when a user enters their initial random number, this spreadsheet is filled in again, creating a unique world. Then, the program models how these data points interact by simulating a daily cycle in which each character has the opportunity to perform an action. The actions in turn are dictated by the traits and relationships of an individual. "A character can not take action to ridicule another character unless he has the" great "trait and his goal is someone who does not like it," says Ryan.
Combine a sufficient number of these data points and you will soon end up with something very complex, Ryan says, and it helps that the program does not just simulate each version of Sheldon County for days or weeks. Simulate centuries "This produces a lot of action, and from this, we can take the most interesting sequences," says Ryan.
He gives the example of an initial project that used similar software to simulate social life in a small city. "In that game, we were always finding ridiculously emerging stories," he says. "In one version, there was a 17-year-old boy whose mom had the favorite restaurant in the city, but then decides to start his own restaurant that becomes more popular than hers!" He had codified how commercial rivalries might work, but never I had expected it to become a family affair. "
You can imagine that argument making a good episode of This American Life.
In many ways, this type of procedure generation is nothing new. It goes back to the 1980s and earlier, says Riedl, where similar techniques were used to generate maps for video games. "Initially it was because the computers of the time did not have much storage space, so that gigantic games could not be sent," he explains. "That meant that it was incumbent upon the first titles, such as Rogue, NetHack, etc., to generate labyrinths in the CPUs using very fast and cheap algorithms."
The utility of these algorithms receded as storage became cheaper, but the procedural approach to generating content has once again become a mainstream as emotion over artificial intelligence has grown.
What is interesting, however, is that modern AI techniques, such as deep neural networks, are not really as suitable for projects as Sheldon County. Ryan says he mainly uses what is sometimes called symbolic AI or, pejoratively, "good old-fashioned intelligence." This approach is less about mining data to look for patterns, such as with deep learning, and more about creating sets of rules and logical instructions that guide a process.
There are some simple reasons why modern AI does not work for tasks like this, says Riedl. In part, techniques such as deep learning are still not good at generating coherent text (even the most advanced chatbots today are based on preprogrammed phrases). And also because older techniques give programmers more control over the output.
Even with these limitations, there are still many things you can do, of course. "One of the most interesting recent examples was the MIT horror story generator, which was interactive," says Riedl. "Then you would write a line, then write a line, you write a line, write a line, it actually becomes this kind of creative writing exercise that allows you to interpret what the program produces."
Using AI creatively means creating new types of entertainment
Riedl's own work explores how modern AI techniques can take on the challenges of storytelling. "I would like to be the person who solves some of these issues to make story generation work better for longer stories," he says. However, it suggests that in the near future, AI will not be able to generate new content for us, be it television programs or podcasts. Instead, it will be the task of a creator to create new types of entertainment.
Alex Champandard, an AI programmer and co-founder of startup Creative.AI, has a similar vision. He says that fully automated entertainment is not the way to go and points out that many creative projects that people claim are a product of AI are based on substantial human vision and supervision. "There is not a machine that has been creative by itself, ultimately it is due to human participation," Champandard tells The Verge.
He suggests that Ryan's project is interesting not necessarily because of the narratives he creates, but because he questions our idea of what a podcast is and what makes one good in the first place. "In the future, I am sure we will have podcasts generated by machines that nobody will listen to, and they will disappear because it will not be worth the electricity," says Champandard. "But these experiments are part of what makes this space fun."
For Ryan, it certainly seems that the sense of experimentation and creation is central to his project. "As far as the future is concerned, I certainly hope generative means will become more conventional," he says, but notes that his own sense of enjoyment comes not from consuming generative means, but from creating them. "The most rewarding thing about working with these simulations is to see these wild emergencies," he says, which is similar to Riedl's argument about MIT's horror story generator. It's just fun because the audience participates.
Maybe that's where the future of generative media resides. Not in the creation of infinite podcasts adapted to each individual, but in turning us all into narrators who create and enjoy our own narratives.