Prime Directive

To ponder, question, and challenge what narrative means in the context of machine learning and artificial intelligence.
To propose and explore a human-machine spectrum for contemporary narrative: human-told, human-interpreted; human-told, machine-interpreted; machine-told, human-interpreted; machine-told, machine-interpreted…

Core Objective

Generate examples and projects that fit towards the right of the spectrum, as hybrids or as “purely” machine-operated.
Attempt to answer: What is the value of generative narrative? Why is it significant? Does it have to be?

Roles and Workflow

Focusing on character-driven (first-person) narrative, we will begin with smaller-sized experiments as proposed:

Annie: video-GAN, neural-storyteller
Reginald: char-RNN, torch-RNN
Tyler: char-RNN, neural-storyteller

Presently, we have approached char-RNN and torch-RNN with training data from Albert Camus, Joseph Conrad, Daniel Keyes, Haruki Murakami. After our individual attempts, we will merge our efforts towards the method with the most potential to generate a larger, more cohesive and fully-realized work.

Questions

Narrative can be very loosely defined, but how do you define the parameters of conventional (or unconventional) stories?
What aspects of this project interests you the most?
What is your preferred mode of narrative?
Can generated narrative replace human stories? If not, what is the role or value of generative narrative?
Who is the author?
How does the absence(?) of human agency in the machine’s writing “process” change the perceived meanings of the work? If the fact that the work was generated by a machine was obscured or hidden, would that elevate or alter perception of the narrative?
What is the ideal “temperature” for generated narrative? (Is it better to appear original with risk for heavy abstraction, or to be legible yet barely deviating from the original input?)

Answers

A narrative should contain sequence, and cause-effect relationships.
Subjects, characters, nouns.
Should be placed in time or have a sense of time?
Work generated by a model that is trained on just one author’s body of work leans too closely to that singular author; work generated by a model that is trained on text from multiple authors begins to enter a more interesting realm of autonomous creation(?)

Proof of Concept

Using char-RNN, trained on Albert Camus’s The Stranger.

Test 1

Temperature: 1
Length: 2000
ming the maty remarked him of prese. One in the red of any matter. But then were of it, to be well, I troup fact, out of evenings and fack her down’s in his cigarettes. Snook and ow any gave a regit, when we were strigging lodder. I gaving are off your hears the wonto: was as a bits of criminal side. Fornow and one cagiince his gloes of gave one. A lawy. When I told make the clome of priconing what though the mit me had not it. I told him I’d never said a day what explain coutt; amout of anstement white thinking any dighty lawed on stayed me read, I walk my broutsly frum the uld’t peess afteir no wonting from breated. Natur no mented at the wandalgo. Raymond had helreving. He had no times, were creiching in the peapes for sept before her here a cups of her, rase being, in the bill wongily and Raymond as he was ensiling his inctaying. “When I all felleat by this it?” But the daze were no lable out of the imins. But overlain own the intent of the deidong the bust be trencual, of newnatking sintly things to morning in she’d been grispless. Then he said that the, however one of the gairtane hone, asked there expect. It would nist seing to mist said the ploselus on the lanch I could down. His questionh that wouse you keopsed something in our dog. “Maraly went of the jury “cour which prose,” he saw out on thairs; the foreathe seath inatker on me, couldn’t wall here how I was setting on my broun. It had my more intentowance. So this would go to make him, though; I gathered he stouse. Wistupped by a might years, as if you to can inmout! you walk to the liver, and no one did how alforried. After a streat on whisen to juy its of precuganted. The old wonent little simperthing, he spoking some. That were said that one deperting: 78 Albert Camus THE STRANGER Inl of daze to see thar, it stayed of doon, the magistrate in the day down, was to swick off his munder but rushed in my arrims didn’t plessely ploscame from the rall roped just him point. It was a s

Test 2

Primetext: “The other day, I “
Temperature: 0.3
Length: 2000
The other day, I thought his way of my mind of the came of the same door and before the street dreps of my counted me any more intention to see the prison of the street and the distance on the relitive still the reliting on the starts of the sun case and the court of the priend of witness. There was one of the scall wrom as in the same down and the sky was aloother and the stairs. It was the prisoner of the same thing to see a sort of me and didn’t any month of the same dead and said the stand that show was a beging to the sunlight still and he said the way and the rattrance, I took the whole or my bedread. And I can glented and the sun was the pressing came and I then a sun in the same from the same thing to see her a country of the boor and the stand of the street drept of my face I had a clees and the rest of she was a small prosent of the start of witness. The doorkeeper a mone of the same proceacidong, I was a bit the prisoner of the sun can of sill beind he could hardly such his and started to him on the water and longed at the sky and looked at the sky with a country talking me to the beach and the prisoner of the same door and the distance of the start the office brought the streetcar all my dog to the death. It was the prosperet of the start of when I was getting of my mind of the read was as to be an off to me to the witness and the street and the bairst was an ear what he was in the other she way on the door. I thought I was some me to a black and had been didnered the doorkeeper a counter me was a sort of streemp and the chope of silence was a string of clumps of that street and the sun garnal what we were the prisoners and the stand of the stands, and I was the sun conded and more of the same thing to see her for the deeper to the prisoner off the first of the street. The sky was a plance of court off the borst of me, and so to have the same the concision of the sun was the striem and the crime was a sort of streember a sort of another course of the hourdest of the su

Test 3

Primetext: “I do not remember “
Temperature: 0.15
Length: 3000
I do not remember my face was a street of whit was no man a street of where I had been to come one of the sun case and the course of the start of where he said to the witness of the start the coffin or not on the same down on the start of my face. I told him I had a longer of the door the street and when I was a black and had a few waiting rather on the state and the court of the execting of the street and the dog on the time with the standss of the start of witness. The street ard asked him if I was the streat of sunder what with a small to the prisoner was a sort of streemp of his sun care too stand to see the same for a strand man’s interest in the prison of it. The sun was an it was a street of part of white of the same door and the street and the dog one was the street and the day was a good three of the death and the street and the dog one was there and the same the old came and the pressing the truem and the rest of the staying and he said the same the prisoner of the same drown and the street and the street and the dog one was there and the sky way the sun was an a short of witness. The street drept of court of the court of the exest to see the sun was the street and the rest of the presiding of the street and the start of which I had a sughers. I realized that the sun was the prisoner of the same door, and the start of where he said, “the police of the start of where he’d said to the wall. I replied that the sune of the stay things on the start of my life was all much at once of the sun was the street and the courtroom and the street and the day when he said the street and the street and the street and the street and the street and the street and the street and the dog one was the prisoner of the same droppers of the day of the court of the execting of the sun came and the street a great of me, and the sun in the courtroom of the street arrads stick at our for a man which I had to know what he said, “the or three you feel you could have came three of the court of the execting me a long round his mother in the start of where he said to me that some of the start and the rest of the countryside a brost of a little whole instace at the same thing to say the street and the dog on the true there the one was the police of the stairs, and the street arm street of whit was down to the start of where the dog one was the prisoner of the same door, and the start of prisoner with my face was all the coffee a strand of my face. It was a street of where of the streetcar and the street and the street and the street and the street and the stand of the sun was the street and the dog on the thing and said the street and the dog was a great of the start of where the streetcars, and the street ard walked back to the street of man cater of the court of the steps of my face when he said, “the cart to the prison say to see that I’d have been going to say to see that they were brought the street and my bedroode which I had a sughe came and the street and the dog on th

Test 4

Primetext: “Mother “
Temperature: 0.01
Mother was a before the street arrads still the sun in the courty of the court offeeper and said the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the street and the s

Final Form

Our final project would ideally be generating stories in real-time, in response to a stimuli (for example: computer vision, object recognition, user-input, voice commands). In addition to human/environmental feedback, multiple machines may interact with each other in the act of storytelling. If not in real-time, the project would likely be either documentation of interactions between multiple neural networks or the result of machine generation – which may or may not be edited/curated/interpreted by human creators.

Real-Time Proposals:

Object Recognition Based Storytelling
Voice Responsive Storytelling

Pre-rendered Proposals:

Generated script driving generated video
Generated video/imagery driving generated text
Generated script for human-made video
Humans-written text training generated text inspiring human-written text

Live Installation with Pre-rendered Media Proposals:

Text to Speech narration of generated novel + word-based image generation / video generation (Neural Storyteller + StackGAN)
Google Deepdream + Neural Storyteller